id	sid	tid	token	lemma	pos
cord-020871-1v6dcmt3	1	1	key	key	NN
cord-020871-1v6dcmt3	1	2	:	:	:
cord-020871-1v6dcmt3	2	1	cord-020871	cord-020871	NNP
cord-020871-1v6dcmt3	2	2	-	-	HYPH
cord-020871-1v6dcmt3	2	3	1v6dcmt3	1v6dcmt3	CD
cord-020871-1v6dcmt3	2	4	authors	author	NNS
cord-020871-1v6dcmt3	2	5	:	:	:
cord-020871-1v6dcmt3	2	6	Papariello	Papariello	NNP
cord-020871-1v6dcmt3	2	7	,	,	,
cord-020871-1v6dcmt3	2	8	Luca	Luca	NNP
cord-020871-1v6dcmt3	2	9	;	;	:
cord-020871-1v6dcmt3	2	10	Bampoulidis	Bampoulidis	NNP
cord-020871-1v6dcmt3	2	11	,	,	,
cord-020871-1v6dcmt3	2	12	Alexandros	Alexandros	NNP
cord-020871-1v6dcmt3	2	13	;	;	:
cord-020871-1v6dcmt3	2	14	Lupu	Lupu	NNP
cord-020871-1v6dcmt3	2	15	,	,	,
cord-020871-1v6dcmt3	2	16	Mihai	Mihai	NNP
cord-020871-1v6dcmt3	2	17	title	title	NN
cord-020871-1v6dcmt3	2	18	:	:	:
cord-020871-1v6dcmt3	3	1	On	on	IN
cord-020871-1v6dcmt3	3	2	the	the	DT
cord-020871-1v6dcmt3	3	3	Replicability	Replicability	NNP
cord-020871-1v6dcmt3	3	4	of	of	IN
cord-020871-1v6dcmt3	3	5	Combining	combine	VBG
cord-020871-1v6dcmt3	3	6	Word	Word	NNP
cord-020871-1v6dcmt3	3	7	Embeddings	Embeddings	NNP
cord-020871-1v6dcmt3	3	8	and	and	CC
cord-020871-1v6dcmt3	3	9	Retrieval	Retrieval	NNP
cord-020871-1v6dcmt3	3	10	Models	Models	NNPS
cord-020871-1v6dcmt3	3	11	date	date	NN
cord-020871-1v6dcmt3	3	12	:	:	:
cord-020871-1v6dcmt3	3	13	2020	2020	CD
cord-020871-1v6dcmt3	3	14	-	-	HYPH
cord-020871-1v6dcmt3	3	15	03	03	CD
cord-020871-1v6dcmt3	3	16	-	-	HYPH
cord-020871-1v6dcmt3	3	17	24	24	CD
cord-020871-1v6dcmt3	3	18	journal	journal	NN
cord-020871-1v6dcmt3	3	19	:	:	:
cord-020871-1v6dcmt3	3	20	Advances	advance	NNS
cord-020871-1v6dcmt3	3	21	in	in	IN
cord-020871-1v6dcmt3	3	22	Information	Information	NNP
cord-020871-1v6dcmt3	3	23	Retrieval	Retrieval	NNP
cord-020871-1v6dcmt3	3	24	DOI	DOI	NNP
cord-020871-1v6dcmt3	3	25	:	:	:
cord-020871-1v6dcmt3	3	26	10.1007/978	10.1007/978	CD
cord-020871-1v6dcmt3	3	27	-	-	HYPH
cord-020871-1v6dcmt3	3	28	3	3	CD
cord-020871-1v6dcmt3	3	29	-	-	HYPH
cord-020871-1v6dcmt3	3	30	030	030	CD
cord-020871-1v6dcmt3	3	31	-	-	HYPH
cord-020871-1v6dcmt3	3	32	45442	45442	CD
cord-020871-1v6dcmt3	3	33	-	-	HYPH
cord-020871-1v6dcmt3	3	34	5_7	5_7	CD
cord-020871-1v6dcmt3	3	35	sha	sha	NNP
cord-020871-1v6dcmt3	3	36	:	:	:
cord-020871-1v6dcmt3	3	37	86513b7187c9f4bfcaf76e989df065eb53a9f871	86513b7187c9f4bfcaf76e989df065eb53a9f871	CD
cord-020871-1v6dcmt3	3	38	doc_id	doc_id	CD
cord-020871-1v6dcmt3	3	39	:	:	:
cord-020871-1v6dcmt3	3	40	20871	20871	CD
cord-020871-1v6dcmt3	3	41	cord_uid	cord_uid	NNS
cord-020871-1v6dcmt3	3	42	:	:	:
cord-020871-1v6dcmt3	4	1	1v6dcmt3	1v6dcmt3	CD
cord-020871-1v6dcmt3	5	1	We	-PRON-	PRP
cord-020871-1v6dcmt3	5	2	replicate	replicate	VBP
cord-020871-1v6dcmt3	5	3	recent	recent	JJ
cord-020871-1v6dcmt3	5	4	experiments	experiment	NNS
cord-020871-1v6dcmt3	5	5	attempting	attempt	VBG
cord-020871-1v6dcmt3	5	6	to	to	TO
cord-020871-1v6dcmt3	5	7	demonstrate	demonstrate	VB
cord-020871-1v6dcmt3	5	8	an	an	DT
cord-020871-1v6dcmt3	5	9	attractive	attractive	JJ
cord-020871-1v6dcmt3	5	10	hypothesis	hypothesis	NN
cord-020871-1v6dcmt3	5	11	about	about	IN
cord-020871-1v6dcmt3	5	12	the	the	DT
cord-020871-1v6dcmt3	5	13	use	use	NN
cord-020871-1v6dcmt3	5	14	of	of	IN
cord-020871-1v6dcmt3	5	15	the	the	DT
cord-020871-1v6dcmt3	5	16	Fisher	Fisher	NNP
cord-020871-1v6dcmt3	5	17	kernel	kernel	NN
cord-020871-1v6dcmt3	5	18	framework	framework	NN
cord-020871-1v6dcmt3	5	19	and	and	CC
cord-020871-1v6dcmt3	5	20	mixture	mixture	NN
cord-020871-1v6dcmt3	5	21	models	model	NNS
cord-020871-1v6dcmt3	5	22	for	for	IN
cord-020871-1v6dcmt3	5	23	aggregating	aggregate	VBG
cord-020871-1v6dcmt3	5	24	word	word	NN
cord-020871-1v6dcmt3	5	25	embeddings	embedding	NNS
cord-020871-1v6dcmt3	5	26	towards	towards	IN
cord-020871-1v6dcmt3	5	27	document	document	NN
cord-020871-1v6dcmt3	5	28	representations	representation	NNS
cord-020871-1v6dcmt3	5	29	and	and	CC
cord-020871-1v6dcmt3	5	30	the	the	DT
cord-020871-1v6dcmt3	5	31	use	use	NN
cord-020871-1v6dcmt3	5	32	of	of	IN
cord-020871-1v6dcmt3	5	33	these	these	DT
cord-020871-1v6dcmt3	5	34	representations	representation	NNS
cord-020871-1v6dcmt3	5	35	in	in	IN
cord-020871-1v6dcmt3	5	36	document	document	NN
cord-020871-1v6dcmt3	5	37	classification	classification	NN
cord-020871-1v6dcmt3	5	38	,	,	,
cord-020871-1v6dcmt3	5	39	clustering	clustering	NN
cord-020871-1v6dcmt3	5	40	,	,	,
cord-020871-1v6dcmt3	5	41	and	and	CC
cord-020871-1v6dcmt3	5	42	retrieval	retrieval	NN
cord-020871-1v6dcmt3	5	43	.	.	.
cord-020871-1v6dcmt3	6	1	Specifically	specifically	RB
cord-020871-1v6dcmt3	6	2	,	,	,
cord-020871-1v6dcmt3	6	3	the	the	DT
cord-020871-1v6dcmt3	6	4	hypothesis	hypothesis	NN
cord-020871-1v6dcmt3	6	5	was	be	VBD
cord-020871-1v6dcmt3	6	6	that	that	IN
cord-020871-1v6dcmt3	6	7	the	the	DT
cord-020871-1v6dcmt3	6	8	use	use	NN
cord-020871-1v6dcmt3	6	9	of	of	IN
cord-020871-1v6dcmt3	6	10	a	a	DT
cord-020871-1v6dcmt3	6	11	mixture	mixture	NN
cord-020871-1v6dcmt3	6	12	model	model	NN
cord-020871-1v6dcmt3	6	13	of	of	IN
cord-020871-1v6dcmt3	6	14	von	von	NNP
cord-020871-1v6dcmt3	6	15	Mises	Mises	NNP
cord-020871-1v6dcmt3	6	16	-	-	HYPH
cord-020871-1v6dcmt3	6	17	Fisher	Fisher	NNP
cord-020871-1v6dcmt3	6	18	(	(	-LRB-
cord-020871-1v6dcmt3	6	19	VMF	VMF	NNP
cord-020871-1v6dcmt3	6	20	)	)	-RRB-
cord-020871-1v6dcmt3	6	21	distributions	distribution	NNS
cord-020871-1v6dcmt3	6	22	instead	instead	RB
cord-020871-1v6dcmt3	6	23	of	of	IN
cord-020871-1v6dcmt3	6	24	Gaussian	gaussian	JJ
cord-020871-1v6dcmt3	6	25	distributions	distribution	NNS
cord-020871-1v6dcmt3	6	26	would	would	MD
cord-020871-1v6dcmt3	6	27	be	be	VB
cord-020871-1v6dcmt3	6	28	beneficial	beneficial	JJ
cord-020871-1v6dcmt3	6	29	because	because	IN
cord-020871-1v6dcmt3	6	30	of	of	IN
cord-020871-1v6dcmt3	6	31	the	the	DT
cord-020871-1v6dcmt3	6	32	focus	focus	NN
cord-020871-1v6dcmt3	6	33	on	on	IN
cord-020871-1v6dcmt3	6	34	cosine	cosine	NN
cord-020871-1v6dcmt3	6	35	distances	distance	NNS
cord-020871-1v6dcmt3	6	36	of	of	IN
cord-020871-1v6dcmt3	6	37	both	both	CC
cord-020871-1v6dcmt3	6	38	VMF	VMF	NNP
cord-020871-1v6dcmt3	6	39	and	and	CC
cord-020871-1v6dcmt3	6	40	the	the	DT
cord-020871-1v6dcmt3	6	41	vector	vector	NN
cord-020871-1v6dcmt3	6	42	space	space	NN
cord-020871-1v6dcmt3	6	43	model	model	NN
cord-020871-1v6dcmt3	6	44	traditionally	traditionally	RB
cord-020871-1v6dcmt3	6	45	used	use	VBN
cord-020871-1v6dcmt3	6	46	in	in	IN
cord-020871-1v6dcmt3	6	47	information	information	NN
cord-020871-1v6dcmt3	6	48	retrieval	retrieval	NN
cord-020871-1v6dcmt3	6	49	.	.	.
cord-020871-1v6dcmt3	7	1	Previous	previous	JJ
cord-020871-1v6dcmt3	7	2	experiments	experiment	NNS
cord-020871-1v6dcmt3	7	3	had	have	VBD
cord-020871-1v6dcmt3	7	4	validated	validate	VBN
cord-020871-1v6dcmt3	7	5	this	this	DT
cord-020871-1v6dcmt3	7	6	hypothesis	hypothesis	NN
cord-020871-1v6dcmt3	7	7	.	.	.
cord-020871-1v6dcmt3	8	1	Our	-PRON-	PRP$
cord-020871-1v6dcmt3	8	2	replication	replication	NN
cord-020871-1v6dcmt3	8	3	was	be	VBD
cord-020871-1v6dcmt3	8	4	not	not	RB
cord-020871-1v6dcmt3	8	5	able	able	JJ
cord-020871-1v6dcmt3	8	6	to	to	TO
cord-020871-1v6dcmt3	8	7	validate	validate	VB
cord-020871-1v6dcmt3	8	8	it	-PRON-	PRP
cord-020871-1v6dcmt3	8	9	,	,	,
cord-020871-1v6dcmt3	8	10	despite	despite	IN
cord-020871-1v6dcmt3	8	11	a	a	DT
cord-020871-1v6dcmt3	8	12	large	large	JJ
cord-020871-1v6dcmt3	8	13	parameter	parameter	NN
cord-020871-1v6dcmt3	8	14	scan	scan	JJ
cord-020871-1v6dcmt3	8	15	space	space	NN
cord-020871-1v6dcmt3	8	16	.	.	.
cord-020871-1v6dcmt3	9	1	The	the	DT
cord-020871-1v6dcmt3	9	2	last	last	JJ
cord-020871-1v6dcmt3	9	3	5	5	CD
cord-020871-1v6dcmt3	9	4	years	year	NNS
cord-020871-1v6dcmt3	9	5	have	have	VBP
cord-020871-1v6dcmt3	9	6	seen	see	VBN
cord-020871-1v6dcmt3	9	7	proof	proof	NN
cord-020871-1v6dcmt3	9	8	that	that	IN
cord-020871-1v6dcmt3	9	9	neural	neural	JJ
cord-020871-1v6dcmt3	9	10	network	network	NN
cord-020871-1v6dcmt3	9	11	-	-	HYPH
cord-020871-1v6dcmt3	9	12	based	base	VBN
cord-020871-1v6dcmt3	9	13	word	word	NN
cord-020871-1v6dcmt3	9	14	embedding	embedding	NN
cord-020871-1v6dcmt3	9	15	models	model	NNS
cord-020871-1v6dcmt3	9	16	provide	provide	VBP
cord-020871-1v6dcmt3	9	17	term	term	NN
cord-020871-1v6dcmt3	9	18	representations	representation	NNS
cord-020871-1v6dcmt3	9	19	that	that	WDT
cord-020871-1v6dcmt3	9	20	are	be	VBP
cord-020871-1v6dcmt3	9	21	a	a	DT
cord-020871-1v6dcmt3	9	22	useful	useful	JJ
cord-020871-1v6dcmt3	9	23	information	information	NN
cord-020871-1v6dcmt3	9	24	source	source	NN
cord-020871-1v6dcmt3	9	25	for	for	IN
cord-020871-1v6dcmt3	9	26	a	a	DT
cord-020871-1v6dcmt3	9	27	variety	variety	NN
cord-020871-1v6dcmt3	9	28	of	of	IN
cord-020871-1v6dcmt3	9	29	tasks	task	NNS
cord-020871-1v6dcmt3	9	30	in	in	IN
cord-020871-1v6dcmt3	9	31	natural	natural	JJ
cord-020871-1v6dcmt3	9	32	language	language	NN
cord-020871-1v6dcmt3	9	33	processing	processing	NN
cord-020871-1v6dcmt3	9	34	.	.	.
cord-020871-1v6dcmt3	10	1	In	in	IN
cord-020871-1v6dcmt3	10	2	information	information	NN
cord-020871-1v6dcmt3	10	3	retrieval	retrieval	NN
cord-020871-1v6dcmt3	10	4	(	(	-LRB-
cord-020871-1v6dcmt3	10	5	IR	IR	NNP
cord-020871-1v6dcmt3	10	6	)	)	-RRB-
cord-020871-1v6dcmt3	10	7	,	,	,
cord-020871-1v6dcmt3	10	8	"	"	``
cord-020871-1v6dcmt3	10	9	traditional	traditional	JJ
cord-020871-1v6dcmt3	10	10	"	"	''
cord-020871-1v6dcmt3	10	11	models	model	NNS
cord-020871-1v6dcmt3	10	12	remain	remain	VBP
cord-020871-1v6dcmt3	10	13	a	a	DT
cord-020871-1v6dcmt3	10	14	high	high	JJ
cord-020871-1v6dcmt3	10	15	baseline	baseline	NN
cord-020871-1v6dcmt3	10	16	to	to	TO
cord-020871-1v6dcmt3	10	17	beat	beat	VB
cord-020871-1v6dcmt3	10	18	,	,	,
cord-020871-1v6dcmt3	10	19	particularly	particularly	RB
cord-020871-1v6dcmt3	10	20	when	when	WRB
cord-020871-1v6dcmt3	10	21	considering	consider	VBG
cord-020871-1v6dcmt3	10	22	efficiency	efficiency	NN
cord-020871-1v6dcmt3	10	23	in	in	IN
cord-020871-1v6dcmt3	10	24	addition	addition	NN
cord-020871-1v6dcmt3	10	25	to	to	TO
cord-020871-1v6dcmt3	10	26	effectiveness	effectiveness	VB
cord-020871-1v6dcmt3	10	27	[	[	-LRB-
cord-020871-1v6dcmt3	10	28	6	6	CD
cord-020871-1v6dcmt3	10	29	]	]	-RRB-
cord-020871-1v6dcmt3	10	30	.	.	.
cord-020871-1v6dcmt3	11	1	Combining	combine	VBG
cord-020871-1v6dcmt3	11	2	the	the	DT
cord-020871-1v6dcmt3	11	3	word	word	NN
cord-020871-1v6dcmt3	11	4	embedding	embed	VBG
cord-020871-1v6dcmt3	11	5	models	model	NNS
cord-020871-1v6dcmt3	11	6	with	with	IN
cord-020871-1v6dcmt3	11	7	the	the	DT
cord-020871-1v6dcmt3	11	8	traditional	traditional	JJ
cord-020871-1v6dcmt3	11	9	IR	IR	NNP
cord-020871-1v6dcmt3	11	10	models	model	NNS
cord-020871-1v6dcmt3	11	11	is	be	VBZ
cord-020871-1v6dcmt3	11	12	therefore	therefore	RB
cord-020871-1v6dcmt3	11	13	very	very	RB
cord-020871-1v6dcmt3	11	14	attractive	attractive	JJ
cord-020871-1v6dcmt3	11	15	and	and	CC
cord-020871-1v6dcmt3	11	16	several	several	JJ
cord-020871-1v6dcmt3	11	17	papers	paper	NNS
cord-020871-1v6dcmt3	11	18	have	have	VBP
cord-020871-1v6dcmt3	11	19	attempted	attempt	VBN
cord-020871-1v6dcmt3	11	20	to	to	TO
cord-020871-1v6dcmt3	11	21	improve	improve	VB
cord-020871-1v6dcmt3	11	22	the	the	DT
cord-020871-1v6dcmt3	11	23	baseline	baseline	NN
cord-020871-1v6dcmt3	11	24	by	by	IN
cord-020871-1v6dcmt3	11	25	adding	add	VBG
cord-020871-1v6dcmt3	11	26	in	in	RP
cord-020871-1v6dcmt3	11	27	,	,	,
cord-020871-1v6dcmt3	11	28	in	in	IN
cord-020871-1v6dcmt3	11	29	a	a	DT
cord-020871-1v6dcmt3	11	30	more	more	RBR
cord-020871-1v6dcmt3	11	31	or	or	CC
cord-020871-1v6dcmt3	11	32	less	less	JJR
cord-020871-1v6dcmt3	11	33	ad	ad	FW
cord-020871-1v6dcmt3	11	34	-	-	HYPH
cord-020871-1v6dcmt3	11	35	hoc	hoc	FW
cord-020871-1v6dcmt3	11	36	fashion	fashion	NN
cord-020871-1v6dcmt3	11	37	,	,	,
cord-020871-1v6dcmt3	11	38	word	word	NN
cord-020871-1v6dcmt3	11	39	-	-	HYPH
cord-020871-1v6dcmt3	11	40	embedding	embed	VBG
cord-020871-1v6dcmt3	11	41	information	information	NN
cord-020871-1v6dcmt3	11	42	.	.	.
cord-020871-1v6dcmt3	12	1	Onal	Onal	NNP
cord-020871-1v6dcmt3	12	2	et	et	NNP
cord-020871-1v6dcmt3	12	3	al	al	NNP
cord-020871-1v6dcmt3	12	4	.	.	.
cord-020871-1v6dcmt3	13	1	[	[	-LRB-
cord-020871-1v6dcmt3	13	2	10	10	CD
cord-020871-1v6dcmt3	13	3	]	]	-RRB-
cord-020871-1v6dcmt3	13	4	summarized	summarize	VBD
cord-020871-1v6dcmt3	13	5	the	the	DT
cord-020871-1v6dcmt3	13	6	various	various	JJ
cord-020871-1v6dcmt3	13	7	developments	development	NNS
cord-020871-1v6dcmt3	13	8	of	of	IN
cord-020871-1v6dcmt3	13	9	the	the	DT
cord-020871-1v6dcmt3	13	10	last	last	JJ
cord-020871-1v6dcmt3	13	11	half	half	JJ
cord-020871-1v6dcmt3	13	12	-	-	HYPH
cord-020871-1v6dcmt3	13	13	decade	decade	NN
cord-020871-1v6dcmt3	13	14	in	in	IN
cord-020871-1v6dcmt3	13	15	the	the	DT
cord-020871-1v6dcmt3	13	16	field	field	NN
cord-020871-1v6dcmt3	13	17	of	of	IN
cord-020871-1v6dcmt3	13	18	neural	neural	JJ
cord-020871-1v6dcmt3	13	19	IR	IR	NNP
cord-020871-1v6dcmt3	13	20	and	and	CC
cord-020871-1v6dcmt3	13	21	group	group	NN
cord-020871-1v6dcmt3	13	22	the	the	DT
cord-020871-1v6dcmt3	13	23	methods	method	NNS
cord-020871-1v6dcmt3	13	24	in	in	IN
cord-020871-1v6dcmt3	13	25	two	two	CD
cord-020871-1v6dcmt3	13	26	categories	category	NNS
cord-020871-1v6dcmt3	13	27	:	:	:
cord-020871-1v6dcmt3	13	28	aggregate	aggregate	JJ
cord-020871-1v6dcmt3	13	29	and	and	CC
cord-020871-1v6dcmt3	13	30	learn	learn	VB
cord-020871-1v6dcmt3	13	31	.	.	.
cord-020871-1v6dcmt3	14	1	The	the	DT
cord-020871-1v6dcmt3	14	2	first	first	JJ
cord-020871-1v6dcmt3	14	3	one	one	NN
cord-020871-1v6dcmt3	14	4	,	,	,
cord-020871-1v6dcmt3	14	5	also	also	RB
cord-020871-1v6dcmt3	14	6	known	know	VBN
cord-020871-1v6dcmt3	14	7	as	as	IN
cord-020871-1v6dcmt3	14	8	compositional	compositional	JJ
cord-020871-1v6dcmt3	14	9	distributional	distributional	JJ
cord-020871-1v6dcmt3	14	10	semantics	semantic	NNS
cord-020871-1v6dcmt3	14	11	,	,	,
cord-020871-1v6dcmt3	14	12	starts	start	VBZ
cord-020871-1v6dcmt3	14	13	from	from	IN
cord-020871-1v6dcmt3	14	14	term	term	NN
cord-020871-1v6dcmt3	14	15	representations	representation	NNS
cord-020871-1v6dcmt3	14	16	and	and	CC
cord-020871-1v6dcmt3	14	17	uses	use	VBZ
cord-020871-1v6dcmt3	14	18	some	some	DT
cord-020871-1v6dcmt3	14	19	function	function	NN
cord-020871-1v6dcmt3	14	20	to	to	TO
cord-020871-1v6dcmt3	14	21	combine	combine	VB
cord-020871-1v6dcmt3	14	22	them	-PRON-	PRP
cord-020871-1v6dcmt3	14	23	into	into	IN
cord-020871-1v6dcmt3	14	24	a	a	DT
cord-020871-1v6dcmt3	14	25	document	document	NN
cord-020871-1v6dcmt3	14	26	representation	representation	NN
cord-020871-1v6dcmt3	14	27	(	(	-LRB-
cord-020871-1v6dcmt3	14	28	a	a	DT
cord-020871-1v6dcmt3	14	29	simple	simple	JJ
cord-020871-1v6dcmt3	14	30	example	example	NN
cord-020871-1v6dcmt3	14	31	is	be	VBZ
cord-020871-1v6dcmt3	14	32	a	a	DT
cord-020871-1v6dcmt3	14	33	weighted	weighted	JJ
cord-020871-1v6dcmt3	14	34	sum	sum	NN
cord-020871-1v6dcmt3	14	35	)	)	-RRB-
cord-020871-1v6dcmt3	14	36	.	.	.
cord-020871-1v6dcmt3	15	1	The	the	DT
cord-020871-1v6dcmt3	15	2	second	second	JJ
cord-020871-1v6dcmt3	15	3	method	method	NN
cord-020871-1v6dcmt3	15	4	uses	use	VBZ
cord-020871-1v6dcmt3	15	5	the	the	DT
cord-020871-1v6dcmt3	15	6	word	word	NN
cord-020871-1v6dcmt3	15	7	embedding	embed	VBG
cord-020871-1v6dcmt3	15	8	as	as	IN
cord-020871-1v6dcmt3	15	9	a	a	DT
cord-020871-1v6dcmt3	15	10	first	first	JJ
cord-020871-1v6dcmt3	15	11	layer	layer	NN
cord-020871-1v6dcmt3	15	12	of	of	IN
cord-020871-1v6dcmt3	15	13	another	another	DT
cord-020871-1v6dcmt3	15	14	neural	neural	JJ
cord-020871-1v6dcmt3	15	15	network	network	NN
cord-020871-1v6dcmt3	15	16	to	to	TO
cord-020871-1v6dcmt3	15	17	output	output	VB
cord-020871-1v6dcmt3	15	18	a	a	DT
cord-020871-1v6dcmt3	15	19	document	document	NN
cord-020871-1v6dcmt3	15	20	representation	representation	NN
cord-020871-1v6dcmt3	15	21	.	.	.
cord-020871-1v6dcmt3	16	1	The	the	DT
cord-020871-1v6dcmt3	16	2	advantage	advantage	NN
cord-020871-1v6dcmt3	16	3	of	of	IN
cord-020871-1v6dcmt3	16	4	the	the	DT
cord-020871-1v6dcmt3	16	5	first	first	JJ
cord-020871-1v6dcmt3	16	6	type	type	NN
cord-020871-1v6dcmt3	16	7	of	of	IN
cord-020871-1v6dcmt3	16	8	methods	method	NNS
cord-020871-1v6dcmt3	16	9	is	be	VBZ
cord-020871-1v6dcmt3	16	10	that	that	IN
cord-020871-1v6dcmt3	16	11	they	-PRON-	PRP
cord-020871-1v6dcmt3	16	12	often	often	RB
cord-020871-1v6dcmt3	16	13	distill	distill	VBP
cord-020871-1v6dcmt3	16	14	down	down	RP
cord-020871-1v6dcmt3	16	15	to	to	IN
cord-020871-1v6dcmt3	16	16	a	a	DT
cord-020871-1v6dcmt3	16	17	linear	linear	JJ
cord-020871-1v6dcmt3	16	18	combination	combination	NN
cord-020871-1v6dcmt3	16	19	(	(	-LRB-
cord-020871-1v6dcmt3	16	20	perhaps	perhaps	RB
cord-020871-1v6dcmt3	16	21	via	via	IN
cord-020871-1v6dcmt3	16	22	a	a	DT
cord-020871-1v6dcmt3	16	23	kernel	kernel	NN
cord-020871-1v6dcmt3	16	24	)	)	-RRB-
cord-020871-1v6dcmt3	16	25	,	,	,
cord-020871-1v6dcmt3	16	26	from	from	IN
cord-020871-1v6dcmt3	16	27	which	which	WDT
cord-020871-1v6dcmt3	16	28	an	an	DT
cord-020871-1v6dcmt3	16	29	explanation	explanation	NN
cord-020871-1v6dcmt3	16	30	about	about	IN
cord-020871-1v6dcmt3	16	31	the	the	DT
cord-020871-1v6dcmt3	16	32	representation	representation	NN
cord-020871-1v6dcmt3	16	33	of	of	IN
cord-020871-1v6dcmt3	16	34	the	the	DT
cord-020871-1v6dcmt3	16	35	document	document	NN
cord-020871-1v6dcmt3	16	36	is	be	VBZ
cord-020871-1v6dcmt3	16	37	easier	easy	JJR
cord-020871-1v6dcmt3	16	38	to	to	TO
cord-020871-1v6dcmt3	16	39	induce	induce	VB
cord-020871-1v6dcmt3	16	40	than	than	IN
cord-020871-1v6dcmt3	16	41	from	from	IN
cord-020871-1v6dcmt3	16	42	the	the	DT
cord-020871-1v6dcmt3	16	43	neural	neural	JJ
cord-020871-1v6dcmt3	16	44	network	network	NN
cord-020871-1v6dcmt3	16	45	layers	layer	NNS
cord-020871-1v6dcmt3	16	46	built	build	VBN
cord-020871-1v6dcmt3	16	47	on	on	IN
cord-020871-1v6dcmt3	16	48	top	top	NN
cord-020871-1v6dcmt3	16	49	of	of	IN
cord-020871-1v6dcmt3	16	50	a	a	DT
cord-020871-1v6dcmt3	16	51	word	word	NN
cord-020871-1v6dcmt3	16	52	embedding	embedding	NN
cord-020871-1v6dcmt3	16	53	.	.	.
cord-020871-1v6dcmt3	17	1	Recently	recently	RB
cord-020871-1v6dcmt3	17	2	,	,	,
cord-020871-1v6dcmt3	17	3	the	the	DT
cord-020871-1v6dcmt3	17	4	issue	issue	NN
cord-020871-1v6dcmt3	17	5	of	of	IN
cord-020871-1v6dcmt3	17	6	explainability	explainability	NN
cord-020871-1v6dcmt3	17	7	in	in	IN
cord-020871-1v6dcmt3	17	8	IR	IR	NNP
cord-020871-1v6dcmt3	17	9	and	and	CC
cord-020871-1v6dcmt3	17	10	recommendation	recommendation	NN
cord-020871-1v6dcmt3	17	11	is	be	VBZ
cord-020871-1v6dcmt3	17	12	generating	generate	VBG
cord-020871-1v6dcmt3	17	13	a	a	DT
cord-020871-1v6dcmt3	17	14	renewed	renew	VBN
cord-020871-1v6dcmt3	17	15	interest	interest	NN
cord-020871-1v6dcmt3	17	16	[	[	-LRB-
cord-020871-1v6dcmt3	17	17	15	15	CD
cord-020871-1v6dcmt3	17	18	]	]	-RRB-
cord-020871-1v6dcmt3	17	19	.	.	.
cord-020871-1v6dcmt3	18	1	In	in	IN
cord-020871-1v6dcmt3	18	2	this	this	DT
cord-020871-1v6dcmt3	18	3	sense	sense	NN
cord-020871-1v6dcmt3	18	4	,	,	,
cord-020871-1v6dcmt3	18	5	Zhang	Zhang	NNP
cord-020871-1v6dcmt3	18	6	et	et	NNP
cord-020871-1v6dcmt3	18	7	al	al	NNP
cord-020871-1v6dcmt3	18	8	.	.	.
cord-020871-1v6dcmt3	19	1	[	[	-LRB-
cord-020871-1v6dcmt3	19	2	14	14	CD
cord-020871-1v6dcmt3	19	3	]	]	-RRB-
cord-020871-1v6dcmt3	19	4	introduced	introduce	VBD
cord-020871-1v6dcmt3	19	5	a	a	DT
cord-020871-1v6dcmt3	19	6	new	new	JJ
cord-020871-1v6dcmt3	19	7	model	model	NN
cord-020871-1v6dcmt3	19	8	for	for	IN
cord-020871-1v6dcmt3	19	9	combining	combine	VBG
cord-020871-1v6dcmt3	19	10	highdimensional	highdimensional	JJ
cord-020871-1v6dcmt3	19	11	vectors	vector	NNS
cord-020871-1v6dcmt3	19	12	,	,	,
cord-020871-1v6dcmt3	19	13	using	use	VBG
cord-020871-1v6dcmt3	19	14	a	a	DT
cord-020871-1v6dcmt3	19	15	mixture	mixture	NN
cord-020871-1v6dcmt3	19	16	model	model	NN
cord-020871-1v6dcmt3	19	17	of	of	IN
cord-020871-1v6dcmt3	19	18	von	von	NNP
cord-020871-1v6dcmt3	19	19	Mises	Mises	NNP
cord-020871-1v6dcmt3	19	20	-	-	HYPH
cord-020871-1v6dcmt3	19	21	Fisher	Fisher	NNP
cord-020871-1v6dcmt3	19	22	(	(	-LRB-
cord-020871-1v6dcmt3	19	23	VMF	VMF	NNP
cord-020871-1v6dcmt3	19	24	)	)	-RRB-
cord-020871-1v6dcmt3	20	1	instead	instead	RB
cord-020871-1v6dcmt3	20	2	of	of	IN
cord-020871-1v6dcmt3	20	3	Gaussian	gaussian	JJ
cord-020871-1v6dcmt3	20	4	distributions	distribution	NNS
cord-020871-1v6dcmt3	20	5	previously	previously	RB
cord-020871-1v6dcmt3	20	6	suggested	suggest	VBN
cord-020871-1v6dcmt3	20	7	by	by	IN
cord-020871-1v6dcmt3	20	8	Clinchant	Clinchant	NNP
cord-020871-1v6dcmt3	20	9	and	and	CC
cord-020871-1v6dcmt3	20	10	Perronnin	Perronnin	NNP
cord-020871-1v6dcmt3	20	11	[	[	-LRB-
cord-020871-1v6dcmt3	20	12	3	3	CD
cord-020871-1v6dcmt3	20	13	]	]	-RRB-
cord-020871-1v6dcmt3	20	14	.	.	.
cord-020871-1v6dcmt3	21	1	This	this	DT
cord-020871-1v6dcmt3	21	2	is	be	VBZ
cord-020871-1v6dcmt3	21	3	an	an	DT
cord-020871-1v6dcmt3	21	4	attractive	attractive	JJ
cord-020871-1v6dcmt3	21	5	hypothesis	hypothesis	NN
cord-020871-1v6dcmt3	21	6	because	because	IN
cord-020871-1v6dcmt3	21	7	the	the	DT
cord-020871-1v6dcmt3	21	8	Gaussian	Gaussian	NNP
cord-020871-1v6dcmt3	21	9	Mixture	Mixture	NNP
cord-020871-1v6dcmt3	21	10	Model	Model	NNP
cord-020871-1v6dcmt3	21	11	(	(	-LRB-
cord-020871-1v6dcmt3	21	12	GMM	GMM	NNP
cord-020871-1v6dcmt3	21	13	)	)	-RRB-
cord-020871-1v6dcmt3	21	14	works	work	VBZ
cord-020871-1v6dcmt3	21	15	on	on	IN
cord-020871-1v6dcmt3	21	16	Euclidean	euclidean	JJ
cord-020871-1v6dcmt3	21	17	distance	distance	NN
cord-020871-1v6dcmt3	21	18	,	,	,
cord-020871-1v6dcmt3	21	19	while	while	IN
cord-020871-1v6dcmt3	21	20	the	the	DT
cord-020871-1v6dcmt3	21	21	mixture	mixture	NN
cord-020871-1v6dcmt3	21	22	of	of	IN
cord-020871-1v6dcmt3	21	23	von	von	NNP
cord-020871-1v6dcmt3	21	24	Mises	Mises	NNP
cord-020871-1v6dcmt3	21	25	-	-	HYPH
cord-020871-1v6dcmt3	21	26	Fisher	Fisher	NNP
cord-020871-1v6dcmt3	21	27	(	(	-LRB-
cord-020871-1v6dcmt3	21	28	moVMF	movmf	NN
cord-020871-1v6dcmt3	21	29	)	)	-RRB-
cord-020871-1v6dcmt3	21	30	model	model	NN
cord-020871-1v6dcmt3	21	31	works	work	VBZ
cord-020871-1v6dcmt3	21	32	on	on	IN
cord-020871-1v6dcmt3	21	33	cosine	cosine	NN
cord-020871-1v6dcmt3	21	34	distances	distance	NNS
cord-020871-1v6dcmt3	21	35	-	-	:
cord-020871-1v6dcmt3	21	36	the	the	DT
cord-020871-1v6dcmt3	21	37	typical	typical	JJ
cord-020871-1v6dcmt3	21	38	distance	distance	NN
cord-020871-1v6dcmt3	21	39	function	function	NN
cord-020871-1v6dcmt3	21	40	in	in	IN
cord-020871-1v6dcmt3	21	41	IR	IR	NNP
cord-020871-1v6dcmt3	21	42	.	.	.
cord-020871-1v6dcmt3	22	1	In	in	IN
cord-020871-1v6dcmt3	22	2	the	the	DT
cord-020871-1v6dcmt3	22	3	following	following	JJ
cord-020871-1v6dcmt3	22	4	sections	section	NNS
cord-020871-1v6dcmt3	22	5	,	,	,
cord-020871-1v6dcmt3	22	6	we	-PRON-	PRP
cord-020871-1v6dcmt3	22	7	set	set	VBD
cord-020871-1v6dcmt3	22	8	up	up	RP
cord-020871-1v6dcmt3	22	9	to	to	TO
cord-020871-1v6dcmt3	22	10	replicate	replicate	VB
cord-020871-1v6dcmt3	22	11	the	the	DT
cord-020871-1v6dcmt3	22	12	experiments	experiment	NNS
cord-020871-1v6dcmt3	22	13	described	describe	VBN
cord-020871-1v6dcmt3	22	14	by	by	IN
cord-020871-1v6dcmt3	22	15	Zhang	Zhang	NNP
cord-020871-1v6dcmt3	22	16	et	et	NNP
cord-020871-1v6dcmt3	22	17	al	al	NNP
cord-020871-1v6dcmt3	22	18	.	.	.
cord-020871-1v6dcmt3	23	1	[	[	-LRB-
cord-020871-1v6dcmt3	23	2	14	14	CD
cord-020871-1v6dcmt3	23	3	]	]	-RRB-
cord-020871-1v6dcmt3	23	4	.	.	.
cord-020871-1v6dcmt3	24	1	They	-PRON-	PRP
cord-020871-1v6dcmt3	24	2	are	be	VBP
cord-020871-1v6dcmt3	24	3	grouped	group	VBN
cord-020871-1v6dcmt3	24	4	in	in	IN
cord-020871-1v6dcmt3	24	5	three	three	CD
cord-020871-1v6dcmt3	24	6	sets	set	NNS
cord-020871-1v6dcmt3	24	7	:	:	:
cord-020871-1v6dcmt3	24	8	classification	classification	NN
cord-020871-1v6dcmt3	24	9	,	,	,
cord-020871-1v6dcmt3	24	10	clustering	clustering	NN
cord-020871-1v6dcmt3	24	11	,	,	,
cord-020871-1v6dcmt3	24	12	and	and	CC
cord-020871-1v6dcmt3	24	13	information	information	NN
cord-020871-1v6dcmt3	24	14	retrieval	retrieval	NN
cord-020871-1v6dcmt3	24	15	,	,	,
cord-020871-1v6dcmt3	24	16	and	and	CC
cord-020871-1v6dcmt3	24	17	compare	compare	VB
cord-020871-1v6dcmt3	24	18	"	"	``
cord-020871-1v6dcmt3	24	19	standard	standard	JJ
cord-020871-1v6dcmt3	24	20	"	"	''
cord-020871-1v6dcmt3	24	21	embedding	embed	VBG
cord-020871-1v6dcmt3	24	22	methods	method	NNS
cord-020871-1v6dcmt3	24	23	with	with	IN
cord-020871-1v6dcmt3	24	24	the	the	DT
cord-020871-1v6dcmt3	24	25	novel	novel	NN
cord-020871-1v6dcmt3	24	26	moVMF	movmf	NN
cord-020871-1v6dcmt3	24	27	representation	representation	NN
cord-020871-1v6dcmt3	24	28	.	.	.
cord-020871-1v6dcmt3	25	1	In	in	IN
cord-020871-1v6dcmt3	25	2	general	general	JJ
cord-020871-1v6dcmt3	25	3	,	,	,
cord-020871-1v6dcmt3	25	4	we	-PRON-	PRP
cord-020871-1v6dcmt3	25	5	follow	follow	VBP
cord-020871-1v6dcmt3	25	6	the	the	DT
cord-020871-1v6dcmt3	25	7	experimental	experimental	JJ
cord-020871-1v6dcmt3	25	8	setup	setup	NN
cord-020871-1v6dcmt3	25	9	of	of	IN
cord-020871-1v6dcmt3	25	10	the	the	DT
cord-020871-1v6dcmt3	25	11	original	original	JJ
cord-020871-1v6dcmt3	25	12	paper	paper	NN
cord-020871-1v6dcmt3	25	13	and	and	CC
cord-020871-1v6dcmt3	25	14	,	,	,
cord-020871-1v6dcmt3	25	15	for	for	IN
cord-020871-1v6dcmt3	25	16	lack	lack	NN
cord-020871-1v6dcmt3	25	17	of	of	IN
cord-020871-1v6dcmt3	25	18	space	space	NN
cord-020871-1v6dcmt3	25	19	,	,	,
cord-020871-1v6dcmt3	25	20	we	-PRON-	PRP
cord-020871-1v6dcmt3	25	21	do	do	VBP
cord-020871-1v6dcmt3	25	22	not	not	RB
cord-020871-1v6dcmt3	25	23	repeat	repeat	VB
cord-020871-1v6dcmt3	25	24	here	here	RB
cord-020871-1v6dcmt3	25	25	many	many	JJ
cord-020871-1v6dcmt3	25	26	details	detail	NNS
cord-020871-1v6dcmt3	25	27	,	,	,
cord-020871-1v6dcmt3	25	28	if	if	IN
cord-020871-1v6dcmt3	25	29	they	-PRON-	PRP
cord-020871-1v6dcmt3	25	30	are	be	VBP
cord-020871-1v6dcmt3	25	31	clearly	clearly	RB
cord-020871-1v6dcmt3	25	32	explained	explain	VBN
cord-020871-1v6dcmt3	25	33	there	there	RB
cord-020871-1v6dcmt3	25	34	.	.	.
cord-020871-1v6dcmt3	26	1	All	all	DT
cord-020871-1v6dcmt3	26	2	experiments	experiment	NNS
cord-020871-1v6dcmt3	26	3	are	be	VBP
cord-020871-1v6dcmt3	26	4	conducted	conduct	VBN
cord-020871-1v6dcmt3	26	5	on	on	IN
cord-020871-1v6dcmt3	26	6	publicly	publicly	RB
cord-020871-1v6dcmt3	26	7	available	available	JJ
cord-020871-1v6dcmt3	26	8	datasets	dataset	NNS
cord-020871-1v6dcmt3	26	9	and	and	CC
cord-020871-1v6dcmt3	26	10	are	be	VBP
cord-020871-1v6dcmt3	26	11	briefly	briefly	RB
cord-020871-1v6dcmt3	26	12	described	describe	VBN
cord-020871-1v6dcmt3	26	13	here	here	RB
cord-020871-1v6dcmt3	26	14	below	below	RB
cord-020871-1v6dcmt3	26	15	.	.	.
cord-020871-1v6dcmt3	27	1	Classification	classification	NN
cord-020871-1v6dcmt3	27	2	.	.	.
cord-020871-1v6dcmt3	28	1	Two	two	CD
cord-020871-1v6dcmt3	28	2	subsets	subset	NNS
cord-020871-1v6dcmt3	28	3	of	of	IN
cord-020871-1v6dcmt3	28	4	the	the	DT
cord-020871-1v6dcmt3	28	5	movie	movie	NN
cord-020871-1v6dcmt3	28	6	review	review	NN
cord-020871-1v6dcmt3	28	7	dataset	dataset	NN
cord-020871-1v6dcmt3	28	8	:	:	:
cord-020871-1v6dcmt3	28	9	(	(	-LRB-
cord-020871-1v6dcmt3	28	10	i	i	NN
cord-020871-1v6dcmt3	28	11	)	)	-RRB-
cord-020871-1v6dcmt3	28	12	the	the	DT
cord-020871-1v6dcmt3	28	13	subjectivity	subjectivity	NN
cord-020871-1v6dcmt3	28	14	dataset	dataset	NN
cord-020871-1v6dcmt3	28	15	(	(	-LRB-
cord-020871-1v6dcmt3	28	16	subj	subj	NNP
cord-020871-1v6dcmt3	28	17	)	)	-RRB-
cord-020871-1v6dcmt3	28	18	[	[	-LRB-
cord-020871-1v6dcmt3	28	19	11	11	CD
cord-020871-1v6dcmt3	28	20	]	]	-RRB-
cord-020871-1v6dcmt3	28	21	;	;	:
cord-020871-1v6dcmt3	28	22	and	and	CC
cord-020871-1v6dcmt3	28	23	(	(	-LRB-
cord-020871-1v6dcmt3	28	24	ii	ii	NN
cord-020871-1v6dcmt3	28	25	)	)	-RRB-
cord-020871-1v6dcmt3	28	26	the	the	DT
cord-020871-1v6dcmt3	28	27	sentence	sentence	NN
cord-020871-1v6dcmt3	28	28	polarity	polarity	NN
cord-020871-1v6dcmt3	28	29	dataset	dataset	NN
cord-020871-1v6dcmt3	28	30	(	(	-LRB-
cord-020871-1v6dcmt3	28	31	sent	send	VBN
cord-020871-1v6dcmt3	28	32	)	)	-RRB-
cord-020871-1v6dcmt3	28	33	[	[	-LRB-
cord-020871-1v6dcmt3	28	34	12	12	CD
cord-020871-1v6dcmt3	28	35	]	]	-RRB-
cord-020871-1v6dcmt3	28	36	.	.	.
cord-020871-1v6dcmt3	29	1	Clustering	cluster	VBG
cord-020871-1v6dcmt3	29	2	.	.	.
cord-020871-1v6dcmt3	30	1	The	the	DT
cord-020871-1v6dcmt3	30	2	20	20	CD
cord-020871-1v6dcmt3	30	3	Newsgroups	newsgroup	NNS
cord-020871-1v6dcmt3	30	4	dataset	dataset	NN
cord-020871-1v6dcmt3	30	5	1	1	CD
cord-020871-1v6dcmt3	30	6	was	be	VBD
cord-020871-1v6dcmt3	30	7	used	use	VBN
cord-020871-1v6dcmt3	30	8	in	in	IN
cord-020871-1v6dcmt3	30	9	the	the	DT
cord-020871-1v6dcmt3	30	10	original	original	JJ
cord-020871-1v6dcmt3	30	11	paper	paper	NN
cord-020871-1v6dcmt3	30	12	,	,	,
cord-020871-1v6dcmt3	30	13	but	but	CC
cord-020871-1v6dcmt3	30	14	the	the	DT
cord-020871-1v6dcmt3	30	15	concrete	concrete	JJ
cord-020871-1v6dcmt3	30	16	version	version	NN
cord-020871-1v6dcmt3	30	17	was	be	VBD
cord-020871-1v6dcmt3	30	18	not	not	RB
cord-020871-1v6dcmt3	30	19	specified	specify	VBN
cord-020871-1v6dcmt3	30	20	.	.	.
cord-020871-1v6dcmt3	31	1	We	-PRON-	PRP
cord-020871-1v6dcmt3	31	2	selected	select	VBD
cord-020871-1v6dcmt3	31	3	the	the	DT
cord-020871-1v6dcmt3	31	4	"	"	``
cord-020871-1v6dcmt3	31	5	bydate	bydate	NNP
cord-020871-1v6dcmt3	31	6	"	"	''
cord-020871-1v6dcmt3	31	7	version	version	NN
cord-020871-1v6dcmt3	31	8	,	,	,
cord-020871-1v6dcmt3	31	9	because	because	IN
cord-020871-1v6dcmt3	31	10	it	-PRON-	PRP
cord-020871-1v6dcmt3	31	11	is	be	VBZ
cord-020871-1v6dcmt3	31	12	,	,	,
cord-020871-1v6dcmt3	31	13	according	accord	VBG
cord-020871-1v6dcmt3	31	14	to	to	IN
cord-020871-1v6dcmt3	31	15	its	-PRON-	PRP$
cord-020871-1v6dcmt3	31	16	creators	creator	NNS
cord-020871-1v6dcmt3	31	17	,	,	,
cord-020871-1v6dcmt3	31	18	the	the	DT
cord-020871-1v6dcmt3	31	19	most	most	RBS
cord-020871-1v6dcmt3	31	20	commonly	commonly	RB
cord-020871-1v6dcmt3	31	21	used	use	VBN
cord-020871-1v6dcmt3	31	22	in	in	IN
cord-020871-1v6dcmt3	31	23	the	the	DT
cord-020871-1v6dcmt3	31	24	literature	literature	NN
cord-020871-1v6dcmt3	31	25	.	.	.
cord-020871-1v6dcmt3	32	1	It	-PRON-	PRP
cord-020871-1v6dcmt3	32	2	is	be	VBZ
cord-020871-1v6dcmt3	32	3	also	also	RB
cord-020871-1v6dcmt3	32	4	the	the	DT
cord-020871-1v6dcmt3	32	5	version	version	NN
cord-020871-1v6dcmt3	32	6	directly	directly	RB
cord-020871-1v6dcmt3	32	7	load	load	NN
cord-020871-1v6dcmt3	32	8	-	-	HYPH
cord-020871-1v6dcmt3	32	9	able	able	JJ
cord-020871-1v6dcmt3	32	10	in	in	IN
cord-020871-1v6dcmt3	32	11	scikit	scikit	NNS
cord-020871-1v6dcmt3	32	12	-	-	:
cord-020871-1v6dcmt3	32	13	learn	learn	VB
cord-020871-1v6dcmt3	32	14	2	2	CD
cord-020871-1v6dcmt3	32	15	,	,	,
cord-020871-1v6dcmt3	32	16	making	make	VBG
cord-020871-1v6dcmt3	32	17	it	-PRON-	PRP
cord-020871-1v6dcmt3	32	18	therefore	therefore	RB
cord-020871-1v6dcmt3	32	19	more	more	RBR
cord-020871-1v6dcmt3	32	20	likely	likely	JJ
cord-020871-1v6dcmt3	32	21	that	that	IN
cord-020871-1v6dcmt3	32	22	the	the	DT
cord-020871-1v6dcmt3	32	23	authors	author	NNS
cord-020871-1v6dcmt3	32	24	had	have	VBD
cord-020871-1v6dcmt3	32	25	used	use	VBN
cord-020871-1v6dcmt3	32	26	this	this	DT
cord-020871-1v6dcmt3	32	27	version	version	NN
cord-020871-1v6dcmt3	32	28	.	.	.
cord-020871-1v6dcmt3	33	1	Retrieval	Retrieval	NNP
cord-020871-1v6dcmt3	33	2	.	.	.
cord-020871-1v6dcmt3	34	1	The	the	DT
cord-020871-1v6dcmt3	34	2	TREC	TREC	NNP
cord-020871-1v6dcmt3	34	3	Robust04	Robust04	NNP
cord-020871-1v6dcmt3	34	4	collection	collection	NN
cord-020871-1v6dcmt3	34	5	[	[	-LRB-
cord-020871-1v6dcmt3	34	6	13	13	CD
cord-020871-1v6dcmt3	34	7	]	]	-RRB-
cord-020871-1v6dcmt3	34	8	.	.	.
cord-020871-1v6dcmt3	35	1	The	the	DT
cord-020871-1v6dcmt3	35	2	methods	method	NNS
cord-020871-1v6dcmt3	35	3	used	use	VBN
cord-020871-1v6dcmt3	35	4	to	to	TO
cord-020871-1v6dcmt3	35	5	generate	generate	VB
cord-020871-1v6dcmt3	35	6	vectors	vector	NNS
cord-020871-1v6dcmt3	35	7	for	for	IN
cord-020871-1v6dcmt3	35	8	terms	term	NNS
cord-020871-1v6dcmt3	35	9	and	and	CC
cord-020871-1v6dcmt3	35	10	documents	document	NNS
cord-020871-1v6dcmt3	35	11	are	be	VBP
cord-020871-1v6dcmt3	35	12	:	:	:
cord-020871-1v6dcmt3	35	13	TF	TF	NNP
cord-020871-1v6dcmt3	35	14	-	-	HYPH
cord-020871-1v6dcmt3	35	15	IDF	IDF	NNP
cord-020871-1v6dcmt3	35	16	.	.	.
cord-020871-1v6dcmt3	36	1	The	the	DT
cord-020871-1v6dcmt3	36	2	basic	basic	JJ
cord-020871-1v6dcmt3	36	3	term	term	NN
cord-020871-1v6dcmt3	36	4	frequency	frequency	NN
cord-020871-1v6dcmt3	36	5	-inverse	-inverse	HYPH
cord-020871-1v6dcmt3	36	6	document	document	NN
cord-020871-1v6dcmt3	36	7	frequency	frequency	NN
cord-020871-1v6dcmt3	36	8	method	method	NN
cord-020871-1v6dcmt3	36	9	[	[	-LRB-
cord-020871-1v6dcmt3	36	10	5	5	CD
cord-020871-1v6dcmt3	36	11	]	]	-RRB-
cord-020871-1v6dcmt3	36	12	.	.	.
cord-020871-1v6dcmt3	37	1	Implemented	implement	VBN
cord-020871-1v6dcmt3	37	2	in	in	IN
cord-020871-1v6dcmt3	37	3	the	the	DT
cord-020871-1v6dcmt3	37	4	scikit	scikit	NNS
cord-020871-1v6dcmt3	37	5	-	-	HYPH
cord-020871-1v6dcmt3	37	6	learn	learn	VB
cord-020871-1v6dcmt3	37	7	library	library	NN
cord-020871-1v6dcmt3	37	8	3	3	CD
cord-020871-1v6dcmt3	37	9	.	.	.
cord-020871-1v6dcmt3	38	1	[	[	-LRB-
cord-020871-1v6dcmt3	38	2	4	4	CD
cord-020871-1v6dcmt3	38	3	]	]	-RRB-
cord-020871-1v6dcmt3	38	4	.	.	.
cord-020871-1v6dcmt3	39	1	LDA	LDA	NNP
cord-020871-1v6dcmt3	39	2	.	.	.
cord-020871-1v6dcmt3	40	1	Latent	Latent	NNP
cord-020871-1v6dcmt3	40	2	Dirichlet	Dirichlet	NNP
cord-020871-1v6dcmt3	40	3	Allocation	Allocation	NNP
cord-020871-1v6dcmt3	40	4	[	[	-LRB-
cord-020871-1v6dcmt3	40	5	2	2	CD
cord-020871-1v6dcmt3	40	6	]	]	-RRB-
cord-020871-1v6dcmt3	40	7	.	.	.
cord-020871-1v6dcmt3	40	8	cBoW.	cbow.	ADD
cord-020871-1v6dcmt3	40	9	Word2vec	word2vec	ADD
cord-020871-1v6dcmt3	41	1	[	[	-LRB-
cord-020871-1v6dcmt3	41	2	9	9	CD
cord-020871-1v6dcmt3	41	3	]	]	-RRB-
cord-020871-1v6dcmt3	41	4	in	in	IN
cord-020871-1v6dcmt3	41	5	the	the	DT
cord-020871-1v6dcmt3	41	6	Continuous	Continuous	NNP
cord-020871-1v6dcmt3	41	7	Bag	Bag	NNP
cord-020871-1v6dcmt3	41	8	-	-	HYPH
cord-020871-1v6dcmt3	41	9	of	of	IN
cord-020871-1v6dcmt3	41	10	-	-	HYPH
cord-020871-1v6dcmt3	41	11	Word	word	NN
cord-020871-1v6dcmt3	41	12	(	(	-LRB-
cord-020871-1v6dcmt3	41	13	cBow	cBow	NNP
cord-020871-1v6dcmt3	41	14	)	)	-RRB-
cord-020871-1v6dcmt3	41	15	architecture	architecture	NN
cord-020871-1v6dcmt3	41	16	.	.	.
cord-020871-1v6dcmt3	42	1	PV	pv	NN
cord-020871-1v6dcmt3	42	2	-	-	HYPH
cord-020871-1v6dcmt3	42	3	DBOW	DBOW	NNP
cord-020871-1v6dcmt3	42	4	/	/	SYM
cord-020871-1v6dcmt3	42	5	DM	dm	NN
cord-020871-1v6dcmt3	42	6	.	.	.
cord-020871-1v6dcmt3	43	1	Paragraph	paragraph	NN
cord-020871-1v6dcmt3	43	2	vector	vector	NN
cord-020871-1v6dcmt3	43	3	(	(	-LRB-
cord-020871-1v6dcmt3	43	4	PV	pv	NN
cord-020871-1v6dcmt3	43	5	)	)	-RRB-
cord-020871-1v6dcmt3	43	6	is	be	VBZ
cord-020871-1v6dcmt3	43	7	a	a	DT
cord-020871-1v6dcmt3	43	8	document	document	NN
cord-020871-1v6dcmt3	43	9	embedding	embed	VBG
cord-020871-1v6dcmt3	43	10	algorithm	algorithm	NN
cord-020871-1v6dcmt3	43	11	that	that	WDT
cord-020871-1v6dcmt3	43	12	builds	build	VBZ
cord-020871-1v6dcmt3	43	13	on	on	IN
cord-020871-1v6dcmt3	43	14	Word2vec	Word2vec	NNP
cord-020871-1v6dcmt3	43	15	.	.	.
cord-020871-1v6dcmt3	44	1	We	-PRON-	PRP
cord-020871-1v6dcmt3	44	2	use	use	VBP
cord-020871-1v6dcmt3	44	3	here	here	RB
cord-020871-1v6dcmt3	44	4	both	both	CC
cord-020871-1v6dcmt3	44	5	its	-PRON-	PRP$
cord-020871-1v6dcmt3	44	6	implementations	implementation	NNS
cord-020871-1v6dcmt3	44	7	:	:	:
cord-020871-1v6dcmt3	44	8	Distributed	distribute	VBN
cord-020871-1v6dcmt3	44	9	Bag	Bag	NNP
cord-020871-1v6dcmt3	44	10	-	-	HYPH
cord-020871-1v6dcmt3	44	11	of	of	IN
cord-020871-1v6dcmt3	44	12	-	-	HYPH
cord-020871-1v6dcmt3	44	13	Words	word	NNS
cord-020871-1v6dcmt3	44	14	(	(	-LRB-
cord-020871-1v6dcmt3	44	15	PV	PV	NNP
cord-020871-1v6dcmt3	44	16	-	-	HYPH
cord-020871-1v6dcmt3	44	17	DBOW	DBOW	NNP
cord-020871-1v6dcmt3	44	18	)	)	-RRB-
cord-020871-1v6dcmt3	44	19	and	and	CC
cord-020871-1v6dcmt3	44	20	Distributed	Distributed	NNP
cord-020871-1v6dcmt3	44	21	Memory	Memory	NNP
cord-020871-1v6dcmt3	44	22	(	(	-LRB-
cord-020871-1v6dcmt3	44	23	PV	pv	NN
cord-020871-1v6dcmt3	44	24	-	-	HYPH
cord-020871-1v6dcmt3	44	25	DM	DM	NNP
cord-020871-1v6dcmt3	44	26	)	)	-RRB-
cord-020871-1v6dcmt3	44	27	[	[	-LRB-
cord-020871-1v6dcmt3	44	28	7	7	CD
cord-020871-1v6dcmt3	44	29	]	]	-RRB-
cord-020871-1v6dcmt3	44	30	.	.	.
cord-020871-1v6dcmt3	45	1	The	the	DT
cord-020871-1v6dcmt3	45	2	LSI	LSI	NNP
cord-020871-1v6dcmt3	45	3	,	,	,
cord-020871-1v6dcmt3	45	4	LDA	LDA	NNP
cord-020871-1v6dcmt3	45	5	,	,	,
cord-020871-1v6dcmt3	45	6	cBoW	cBoW	NNP
cord-020871-1v6dcmt3	45	7	,	,	,
cord-020871-1v6dcmt3	45	8	and	and	CC
cord-020871-1v6dcmt3	45	9	PV	pv	NN
cord-020871-1v6dcmt3	45	10	implementations	implementation	NNS
cord-020871-1v6dcmt3	45	11	are	be	VBP
cord-020871-1v6dcmt3	45	12	available	available	JJ
cord-020871-1v6dcmt3	45	13	in	in	IN
cord-020871-1v6dcmt3	45	14	the	the	DT
cord-020871-1v6dcmt3	45	15	gensim	gensim	NNP
cord-020871-1v6dcmt3	45	16	library	library	NNP
cord-020871-1v6dcmt3	45	17	4	4	CD
cord-020871-1v6dcmt3	45	18	.	.	.
cord-020871-1v6dcmt3	46	1	The	the	DT
cord-020871-1v6dcmt3	46	2	FK	FK	NNP
cord-020871-1v6dcmt3	46	3	framework	framework	NN
cord-020871-1v6dcmt3	46	4	offers	offer	VBZ
cord-020871-1v6dcmt3	46	5	the	the	DT
cord-020871-1v6dcmt3	46	6	option	option	NN
cord-020871-1v6dcmt3	46	7	to	to	TO
cord-020871-1v6dcmt3	46	8	aggregate	aggregate	VB
cord-020871-1v6dcmt3	46	9	word	word	NN
cord-020871-1v6dcmt3	46	10	embeddings	embedding	NNS
cord-020871-1v6dcmt3	46	11	to	to	TO
cord-020871-1v6dcmt3	46	12	obtain	obtain	VB
cord-020871-1v6dcmt3	46	13	fixed	fix	VBN
cord-020871-1v6dcmt3	46	14	-	-	HYPH
cord-020871-1v6dcmt3	46	15	length	length	NN
cord-020871-1v6dcmt3	46	16	representations	representation	NNS
cord-020871-1v6dcmt3	46	17	of	of	IN
cord-020871-1v6dcmt3	46	18	documents	document	NNS
cord-020871-1v6dcmt3	46	19	.	.	.
cord-020871-1v6dcmt3	47	1	We	-PRON-	PRP
cord-020871-1v6dcmt3	47	2	use	use	VBP
cord-020871-1v6dcmt3	47	3	Fisher	Fisher	NNP
cord-020871-1v6dcmt3	47	4	vectors	vector	NNS
cord-020871-1v6dcmt3	47	5	(	(	-LRB-
cord-020871-1v6dcmt3	47	6	FV	FV	NNP
cord-020871-1v6dcmt3	47	7	)	)	-RRB-
cord-020871-1v6dcmt3	47	8	based	base	VBN
cord-020871-1v6dcmt3	47	9	on	on	IN
cord-020871-1v6dcmt3	47	10	(	(	-LRB-
cord-020871-1v6dcmt3	47	11	i	i	NN
cord-020871-1v6dcmt3	47	12	)	)	-RRB-
cord-020871-1v6dcmt3	47	13	a	a	DT
cord-020871-1v6dcmt3	47	14	Gaussian	gaussian	JJ
cord-020871-1v6dcmt3	47	15	mixture	mixture	NN
cord-020871-1v6dcmt3	47	16	model	model	NN
cord-020871-1v6dcmt3	47	17	(	(	-LRB-
cord-020871-1v6dcmt3	47	18	FV	FV	NNP
cord-020871-1v6dcmt3	47	19	-	-	HYPH
cord-020871-1v6dcmt3	47	20	GMM	GMM	NNP
cord-020871-1v6dcmt3	47	21	)	)	-RRB-
cord-020871-1v6dcmt3	47	22	and	and	CC
cord-020871-1v6dcmt3	47	23	(	(	-LRB-
cord-020871-1v6dcmt3	47	24	ii	ii	NN
cord-020871-1v6dcmt3	47	25	)	)	-RRB-
cord-020871-1v6dcmt3	47	26	a	a	DT
cord-020871-1v6dcmt3	47	27	mixture	mixture	NN
cord-020871-1v6dcmt3	47	28	of	of	IN
cord-020871-1v6dcmt3	47	29	von	von	NNP
cord-020871-1v6dcmt3	47	30	Mises	Mises	NNP
cord-020871-1v6dcmt3	47	31	-	-	HYPH
cord-020871-1v6dcmt3	47	32	Fisher	Fisher	NNP
cord-020871-1v6dcmt3	47	33	distributions	distribution	NNS
cord-020871-1v6dcmt3	47	34	(	(	-LRB-
cord-020871-1v6dcmt3	47	35	FV	fv	NN
cord-020871-1v6dcmt3	47	36	-	-	HYPH
cord-020871-1v6dcmt3	47	37	moVMF	movmf	NN
cord-020871-1v6dcmt3	47	38	)	)	-RRB-
cord-020871-1v6dcmt3	47	39	[	[	-LRB-
cord-020871-1v6dcmt3	47	40	1	1	CD
cord-020871-1v6dcmt3	47	41	]	]	-RRB-
cord-020871-1v6dcmt3	47	42	.	.	.
cord-020871-1v6dcmt3	48	1	We	-PRON-	PRP
cord-020871-1v6dcmt3	48	2	first	first	RB
cord-020871-1v6dcmt3	48	3	fit	fit	VBP
cord-020871-1v6dcmt3	48	4	(	(	-LRB-
cord-020871-1v6dcmt3	48	5	i	i	NN
cord-020871-1v6dcmt3	48	6	)	)	-RRB-
cord-020871-1v6dcmt3	49	1	a	a	DT
cord-020871-1v6dcmt3	49	2	GMM	GMM	NNP
cord-020871-1v6dcmt3	49	3	and	and	CC
cord-020871-1v6dcmt3	49	4	(	(	-LRB-
cord-020871-1v6dcmt3	49	5	ii	ii	NN
cord-020871-1v6dcmt3	49	6	)	)	-RRB-
cord-020871-1v6dcmt3	49	7	a	a	DT
cord-020871-1v6dcmt3	49	8	moVMF	movmf	NN
cord-020871-1v6dcmt3	49	9	model	model	NN
cord-020871-1v6dcmt3	49	10	on	on	IN
cord-020871-1v6dcmt3	49	11	previously	previously	RB
cord-020871-1v6dcmt3	49	12	learnt	learn	VBN
cord-020871-1v6dcmt3	49	13	continuous	continuous	JJ
cord-020871-1v6dcmt3	49	14	word	word	NN
cord-020871-1v6dcmt3	49	15	embeddings	embedding	NNS
cord-020871-1v6dcmt3	49	16	.	.	.
cord-020871-1v6dcmt3	50	1	The	the	DT
cord-020871-1v6dcmt3	50	2	fixed	fix	VBN
cord-020871-1v6dcmt3	50	3	-	-	HYPH
cord-020871-1v6dcmt3	50	4	length	length	NN
cord-020871-1v6dcmt3	50	5	representation	representation	NN
cord-020871-1v6dcmt3	50	6	of	of	IN
cord-020871-1v6dcmt3	50	7	a	a	DT
cord-020871-1v6dcmt3	50	8	document	document	NN
cord-020871-1v6dcmt3	50	9	X	x	NN
cord-020871-1v6dcmt3	50	10	containing	contain	VBG
cord-020871-1v6dcmt3	50	11	T	T	NNP
cord-020871-1v6dcmt3	50	12	words	word	NNS
cord-020871-1v6dcmt3	51	1	w	w	NNP
cord-020871-1v6dcmt3	51	2	i	i	PRP
cord-020871-1v6dcmt3	52	1	-expressed	-expressed	JJ
cord-020871-1v6dcmt3	53	1	as	as	IN
cord-020871-1v6dcmt3	53	2	where	where	WRB
cord-020871-1v6dcmt3	53	3	K	K	NNP
cord-020871-1v6dcmt3	53	4	is	be	VBZ
cord-020871-1v6dcmt3	53	5	the	the	DT
cord-020871-1v6dcmt3	53	6	number	number	NN
cord-020871-1v6dcmt3	53	7	of	of	IN
cord-020871-1v6dcmt3	53	8	mixture	mixture	NN
cord-020871-1v6dcmt3	53	9	components	component	NNS
cord-020871-1v6dcmt3	53	10	.	.	.
cord-020871-1v6dcmt3	54	1	The	the	DT
cord-020871-1v6dcmt3	54	2	vectors	vector	NNS
cord-020871-1v6dcmt3	54	3	G	g	NN
cord-020871-1v6dcmt3	54	4	X	x	NN
cord-020871-1v6dcmt3	54	5	i	i	NN
cord-020871-1v6dcmt3	54	6	,	,	,
cord-020871-1v6dcmt3	54	7	having	have	VBG
cord-020871-1v6dcmt3	54	8	the	the	DT
cord-020871-1v6dcmt3	54	9	dimension	dimension	NN
cord-020871-1v6dcmt3	54	10	(	(	-LRB-
cord-020871-1v6dcmt3	54	11	d	d	NN
cord-020871-1v6dcmt3	54	12	)	)	-RRB-
cord-020871-1v6dcmt3	54	13	of	of	IN
cord-020871-1v6dcmt3	54	14	the	the	DT
cord-020871-1v6dcmt3	54	15	word	word	NN
cord-020871-1v6dcmt3	54	16	vectors	vector	NNS
cord-020871-1v6dcmt3	54	17	E	E	NNP
cord-020871-1v6dcmt3	54	18	wi	wi	NNP
cord-020871-1v6dcmt3	54	19	,	,	,
cord-020871-1v6dcmt3	54	20	are	be	VBP
cord-020871-1v6dcmt3	54	21	explicitly	explicitly	RB
cord-020871-1v6dcmt3	54	22	given	give	VBN
cord-020871-1v6dcmt3	54	23	by	by	IN
cord-020871-1v6dcmt3	54	24	[	[	-LRB-
cord-020871-1v6dcmt3	54	25	3	3	CD
cord-020871-1v6dcmt3	54	26	,	,	,
cord-020871-1v6dcmt3	54	27	14	14	CD
cord-020871-1v6dcmt3	54	28	]	]	-RRB-
cord-020871-1v6dcmt3	54	29	:	:	:
cord-020871-1v6dcmt3	54	30	where	where	WRB
cord-020871-1v6dcmt3	54	31	ω	ω	NN
cord-020871-1v6dcmt3	54	32	i	i	PRP
cord-020871-1v6dcmt3	54	33	are	be	VBP
cord-020871-1v6dcmt3	54	34	the	the	DT
cord-020871-1v6dcmt3	54	35	mixture	mixture	NN
cord-020871-1v6dcmt3	54	36	weights	weight	NNS
cord-020871-1v6dcmt3	54	37	,	,	,
cord-020871-1v6dcmt3	54	38	γ	γ	NNP
cord-020871-1v6dcmt3	54	39	t	t	NN
cord-020871-1v6dcmt3	54	40	(	(	-LRB-
cord-020871-1v6dcmt3	54	41	i	i	NN
cord-020871-1v6dcmt3	54	42	)	)	-RRB-
cord-020871-1v6dcmt3	54	43	=	=	SYM
cord-020871-1v6dcmt3	54	44	p(i|x	p(i|x	NNP
cord-020871-1v6dcmt3	54	45	t	t	NN
cord-020871-1v6dcmt3	54	46	)	)	-RRB-
cord-020871-1v6dcmt3	54	47	is	be	VBZ
cord-020871-1v6dcmt3	54	48	the	the	DT
cord-020871-1v6dcmt3	54	49	soft	soft	JJ
cord-020871-1v6dcmt3	54	50	assignment	assignment	NN
cord-020871-1v6dcmt3	54	51	of	of	IN
cord-020871-1v6dcmt3	54	52	x	x	SYM
cord-020871-1v6dcmt3	54	53	t	t	NN
cord-020871-1v6dcmt3	55	1	to	to	IN
cord-020871-1v6dcmt3	55	2	(	(	-LRB-
cord-020871-1v6dcmt3	55	3	i	i	NN
cord-020871-1v6dcmt3	55	4	)	)	-RRB-
cord-020871-1v6dcmt3	55	5	Gaussian	Gaussian	NNP
cord-020871-1v6dcmt3	55	6	and	and	CC
cord-020871-1v6dcmt3	55	7	(	(	-LRB-
cord-020871-1v6dcmt3	55	8	ii	ii	NN
cord-020871-1v6dcmt3	55	9	)	)	-RRB-
cord-020871-1v6dcmt3	56	1	VMF	VMF	NNP
cord-020871-1v6dcmt3	56	2	distribution	distribution	NN
cord-020871-1v6dcmt3	56	3	i	i	PRP
cord-020871-1v6dcmt3	56	4	,	,	,
cord-020871-1v6dcmt3	56	5	and	and	CC
cord-020871-1v6dcmt3	56	6	σ	σ	NNP
cord-020871-1v6dcmt3	56	7	2	2	CD
cord-020871-1v6dcmt3	56	8	i	i	NNP
cord-020871-1v6dcmt3	56	9	=	=	SYM
cord-020871-1v6dcmt3	56	10	diag(Σ	diag(Σ	NNS
cord-020871-1v6dcmt3	56	11	i	i	PRP
cord-020871-1v6dcmt3	56	12	)	)	-RRB-
cord-020871-1v6dcmt3	56	13	,	,	,
cord-020871-1v6dcmt3	56	14	with	with	IN
cord-020871-1v6dcmt3	56	15	Σ	Σ	NNP
cord-020871-1v6dcmt3	56	16	i	i	PRP
cord-020871-1v6dcmt3	56	17	the	the	DT
cord-020871-1v6dcmt3	56	18	covariance	covariance	NN
cord-020871-1v6dcmt3	56	19	matrix	matrix	NN
cord-020871-1v6dcmt3	56	20	of	of	IN
cord-020871-1v6dcmt3	56	21	Gaussian	Gaussian	NNP
cord-020871-1v6dcmt3	56	22	i.	i.	NN
cord-020871-1v6dcmt3	57	1	In	in	IN
cord-020871-1v6dcmt3	57	2	(	(	-LRB-
cord-020871-1v6dcmt3	57	3	i	i	NN
cord-020871-1v6dcmt3	57	4	)	)	-RRB-
cord-020871-1v6dcmt3	57	5	,	,	,
cord-020871-1v6dcmt3	57	6	σ	σ	NNP
cord-020871-1v6dcmt3	58	1	i	i	PRP
cord-020871-1v6dcmt3	58	2	refers	refer	VBZ
cord-020871-1v6dcmt3	58	3	to	to	IN
cord-020871-1v6dcmt3	58	4	the	the	DT
cord-020871-1v6dcmt3	58	5	mean	mean	JJ
cord-020871-1v6dcmt3	58	6	vector	vector	NN
cord-020871-1v6dcmt3	58	7	;	;	:
cord-020871-1v6dcmt3	58	8	in	in	IN
cord-020871-1v6dcmt3	58	9	(	(	-LRB-
cord-020871-1v6dcmt3	58	10	ii	ii	NN
cord-020871-1v6dcmt3	58	11	)	)	-RRB-
cord-020871-1v6dcmt3	58	12	it	-PRON-	PRP
cord-020871-1v6dcmt3	58	13	indicates	indicate	VBZ
cord-020871-1v6dcmt3	58	14	the	the	DT
cord-020871-1v6dcmt3	58	15	mean	mean	JJ
cord-020871-1v6dcmt3	58	16	direction	direction	NN
cord-020871-1v6dcmt3	58	17	and	and	CC
cord-020871-1v6dcmt3	58	18	κ	κ	NN
cord-020871-1v6dcmt3	59	1	i	i	PRP
cord-020871-1v6dcmt3	59	2	is	be	VBZ
cord-020871-1v6dcmt3	59	3	the	the	DT
cord-020871-1v6dcmt3	59	4	concentration	concentration	NN
cord-020871-1v6dcmt3	59	5	parameter	parameter	NN
cord-020871-1v6dcmt3	59	6	.	.	.
cord-020871-1v6dcmt3	60	1	We	-PRON-	PRP
cord-020871-1v6dcmt3	60	2	implement	implement	VBP
cord-020871-1v6dcmt3	60	3	the	the	DT
cord-020871-1v6dcmt3	60	4	FK	FK	NNP
cord-020871-1v6dcmt3	60	5	-	-	HYPH
cord-020871-1v6dcmt3	60	6	based	base	VBN
cord-020871-1v6dcmt3	60	7	algorithms	algorithm	NNS
cord-020871-1v6dcmt3	60	8	by	by	IN
cord-020871-1v6dcmt3	60	9	ourselves	-PRON-	PRP
cord-020871-1v6dcmt3	60	10	,	,	,
cord-020871-1v6dcmt3	60	11	with	with	IN
cord-020871-1v6dcmt3	60	12	the	the	DT
cord-020871-1v6dcmt3	60	13	help	help	NN
cord-020871-1v6dcmt3	60	14	of	of	IN
cord-020871-1v6dcmt3	60	15	the	the	DT
cord-020871-1v6dcmt3	60	16	scikit	scikit	NNS
cord-020871-1v6dcmt3	60	17	-	-	HYPH
cord-020871-1v6dcmt3	60	18	learn	learn	VB
cord-020871-1v6dcmt3	60	19	library	library	NN
cord-020871-1v6dcmt3	60	20	for	for	IN
cord-020871-1v6dcmt3	60	21	fitting	fit	VBG
cord-020871-1v6dcmt3	60	22	a	a	DT
cord-020871-1v6dcmt3	60	23	mixture	mixture	NN
cord-020871-1v6dcmt3	60	24	of	of	IN
cord-020871-1v6dcmt3	60	25	Gaussian	gaussian	JJ
cord-020871-1v6dcmt3	60	26	models	model	NNS
cord-020871-1v6dcmt3	60	27	and	and	CC
cord-020871-1v6dcmt3	60	28	of	of	IN
cord-020871-1v6dcmt3	60	29	the	the	DT
cord-020871-1v6dcmt3	60	30	Spherecluster	Spherecluster	NNP
cord-020871-1v6dcmt3	60	31	package	package	NN
cord-020871-1v6dcmt3	60	32	5	5	CD
cord-020871-1v6dcmt3	60	33	for	for	IN
cord-020871-1v6dcmt3	60	34	fitting	fit	VBG
cord-020871-1v6dcmt3	60	35	a	a	DT
cord-020871-1v6dcmt3	60	36	mixture	mixture	NN
cord-020871-1v6dcmt3	60	37	of	of	IN
cord-020871-1v6dcmt3	60	38	von	von	NNP
cord-020871-1v6dcmt3	60	39	Mises	Mises	NNP
cord-020871-1v6dcmt3	60	40	-	-	HYPH
cord-020871-1v6dcmt3	60	41	Fisher	Fisher	NNP
cord-020871-1v6dcmt3	60	42	distributions	distribution	NNS
cord-020871-1v6dcmt3	60	43	to	to	IN
cord-020871-1v6dcmt3	60	44	our	-PRON-	PRP$
cord-020871-1v6dcmt3	60	45	data	datum	NNS
cord-020871-1v6dcmt3	60	46	.	.	.
cord-020871-1v6dcmt3	61	1	The	the	DT
cord-020871-1v6dcmt3	61	2	implementation	implementation	NN
cord-020871-1v6dcmt3	61	3	details	detail	NNS
cord-020871-1v6dcmt3	61	4	of	of	IN
cord-020871-1v6dcmt3	61	5	each	each	DT
cord-020871-1v6dcmt3	61	6	algorithm	algorithm	NN
cord-020871-1v6dcmt3	61	7	are	be	VBP
cord-020871-1v6dcmt3	61	8	described	describe	VBN
cord-020871-1v6dcmt3	61	9	in	in	IN
cord-020871-1v6dcmt3	61	10	what	what	WP
cord-020871-1v6dcmt3	61	11	follows	follow	VBZ
cord-020871-1v6dcmt3	61	12	.	.	.
cord-020871-1v6dcmt3	62	1	Each	each	DT
cord-020871-1v6dcmt3	62	2	of	of	IN
cord-020871-1v6dcmt3	62	3	the	the	DT
cord-020871-1v6dcmt3	62	4	following	follow	VBG
cord-020871-1v6dcmt3	62	5	experiments	experiment	NNS
cord-020871-1v6dcmt3	62	6	is	be	VBZ
cord-020871-1v6dcmt3	62	7	conceptually	conceptually	RB
cord-020871-1v6dcmt3	62	8	divided	divide	VBN
cord-020871-1v6dcmt3	62	9	in	in	IN
cord-020871-1v6dcmt3	62	10	three	three	CD
cord-020871-1v6dcmt3	62	11	phases	phase	NNS
cord-020871-1v6dcmt3	62	12	.	.	.
cord-020871-1v6dcmt3	63	1	First	first	RB
cord-020871-1v6dcmt3	63	2	,	,	,
cord-020871-1v6dcmt3	63	3	text	text	NN
cord-020871-1v6dcmt3	63	4	processing	processing	NN
cord-020871-1v6dcmt3	63	5	(	(	-LRB-
cord-020871-1v6dcmt3	63	6	e.g.	e.g.	RB
cord-020871-1v6dcmt3	63	7	tokenisation	tokenisation	NN
cord-020871-1v6dcmt3	63	8	)	)	-RRB-
cord-020871-1v6dcmt3	63	9	;	;	:
cord-020871-1v6dcmt3	63	10	second	second	RB
cord-020871-1v6dcmt3	63	11	,	,	,
cord-020871-1v6dcmt3	63	12	creating	create	VBG
cord-020871-1v6dcmt3	63	13	a	a	DT
cord-020871-1v6dcmt3	63	14	fixed	fix	VBN
cord-020871-1v6dcmt3	63	15	-	-	HYPH
cord-020871-1v6dcmt3	63	16	length	length	NN
cord-020871-1v6dcmt3	63	17	vector	vector	NN
cord-020871-1v6dcmt3	63	18	representation	representation	NN
cord-020871-1v6dcmt3	63	19	for	for	IN
cord-020871-1v6dcmt3	63	20	every	every	DT
cord-020871-1v6dcmt3	63	21	document	document	NN
cord-020871-1v6dcmt3	63	22	;	;	:
cord-020871-1v6dcmt3	63	23	finally	finally	RB
cord-020871-1v6dcmt3	63	24	,	,	,
cord-020871-1v6dcmt3	63	25	the	the	DT
cord-020871-1v6dcmt3	63	26	third	third	JJ
cord-020871-1v6dcmt3	63	27	phase	phase	NN
cord-020871-1v6dcmt3	63	28	is	be	VBZ
cord-020871-1v6dcmt3	63	29	determined	determine	VBN
cord-020871-1v6dcmt3	63	30	by	by	IN
cord-020871-1v6dcmt3	63	31	the	the	DT
cord-020871-1v6dcmt3	63	32	goal	goal	NN
cord-020871-1v6dcmt3	63	33	to	to	TO
cord-020871-1v6dcmt3	63	34	be	be	VB
cord-020871-1v6dcmt3	63	35	achieved	achieve	VBN
cord-020871-1v6dcmt3	63	36	,	,	,
cord-020871-1v6dcmt3	63	37	i.e.	i.e.	FW
cord-020871-1v6dcmt3	63	38	classification	classification	NN
cord-020871-1v6dcmt3	63	39	,	,	,
cord-020871-1v6dcmt3	63	40	clustering	clustering	NN
cord-020871-1v6dcmt3	63	41	,	,	,
cord-020871-1v6dcmt3	63	42	and	and	CC
cord-020871-1v6dcmt3	63	43	retrieval	retrieval	NN
cord-020871-1v6dcmt3	63	44	.	.	.
cord-020871-1v6dcmt3	64	1	For	for	IN
cord-020871-1v6dcmt3	64	2	the	the	DT
cord-020871-1v6dcmt3	64	3	first	first	JJ
cord-020871-1v6dcmt3	64	4	phase	phase	NN
cord-020871-1v6dcmt3	64	5	the	the	DT
cord-020871-1v6dcmt3	64	6	same	same	JJ
cord-020871-1v6dcmt3	64	7	pre	pre	JJ
cord-020871-1v6dcmt3	64	8	-	-	JJ
cord-020871-1v6dcmt3	64	9	processing	processing	NN
cord-020871-1v6dcmt3	64	10	is	be	VBZ
cord-020871-1v6dcmt3	64	11	applied	apply	VBN
cord-020871-1v6dcmt3	64	12	to	to	IN
cord-020871-1v6dcmt3	64	13	all	all	DT
cord-020871-1v6dcmt3	64	14	datasets	dataset	NNS
cord-020871-1v6dcmt3	64	15	.	.	.
cord-020871-1v6dcmt3	65	1	In	in	IN
cord-020871-1v6dcmt3	65	2	the	the	DT
cord-020871-1v6dcmt3	65	3	original	original	JJ
cord-020871-1v6dcmt3	65	4	paper	paper	NN
cord-020871-1v6dcmt3	65	5	,	,	,
cord-020871-1v6dcmt3	65	6	this	this	DT
cord-020871-1v6dcmt3	65	7	phase	phase	NN
cord-020871-1v6dcmt3	65	8	was	be	VBD
cord-020871-1v6dcmt3	65	9	only	only	RB
cord-020871-1v6dcmt3	65	10	briefly	briefly	RB
cord-020871-1v6dcmt3	65	11	described	describe	VBN
cord-020871-1v6dcmt3	65	12	as	as	IN
cord-020871-1v6dcmt3	65	13	tokenisation	tokenisation	NN
cord-020871-1v6dcmt3	65	14	and	and	CC
cord-020871-1v6dcmt3	65	15	stopword	stopword	NN
cord-020871-1v6dcmt3	65	16	removal	removal	NN
cord-020871-1v6dcmt3	65	17	.	.	.
cord-020871-1v6dcmt3	66	1	It	-PRON-	PRP
cord-020871-1v6dcmt3	66	2	is	be	VBZ
cord-020871-1v6dcmt3	66	3	not	not	RB
cord-020871-1v6dcmt3	66	4	given	give	VBN
cord-020871-1v6dcmt3	66	5	what	what	WP
cord-020871-1v6dcmt3	66	6	tokeniser	tokeniser	NNP
cord-020871-1v6dcmt3	66	7	,	,	,
cord-020871-1v6dcmt3	66	8	linguistic	linguistic	JJ
cord-020871-1v6dcmt3	66	9	filters	filter	NNS
cord-020871-1v6dcmt3	66	10	(	(	-LRB-
cord-020871-1v6dcmt3	66	11	stemming	stem	VBG
cord-020871-1v6dcmt3	66	12	,	,	,
cord-020871-1v6dcmt3	66	13	lemmatisation	lemmatisation	NN
cord-020871-1v6dcmt3	66	14	,	,	,
cord-020871-1v6dcmt3	66	15	etc	etc	FW
cord-020871-1v6dcmt3	66	16	.	.	.
cord-020871-1v6dcmt3	66	17	)	)	-RRB-
cord-020871-1v6dcmt3	66	18	,	,	,
cord-020871-1v6dcmt3	66	19	or	or	CC
cord-020871-1v6dcmt3	66	20	stop	stop	VB
cord-020871-1v6dcmt3	66	21	word	word	NN
cord-020871-1v6dcmt3	66	22	list	list	NN
cord-020871-1v6dcmt3	66	23	were	be	VBD
cord-020871-1v6dcmt3	66	24	used	use	VBN
cord-020871-1v6dcmt3	66	25	.	.	.
cord-020871-1v6dcmt3	67	1	Knowing	know	VBG
cord-020871-1v6dcmt3	67	2	that	that	IN
cord-020871-1v6dcmt3	67	3	the	the	DT
cord-020871-1v6dcmt3	67	4	gensim	gensim	NNP
cord-020871-1v6dcmt3	67	5	library	library	NNP
cord-020871-1v6dcmt3	67	6	was	be	VBD
cord-020871-1v6dcmt3	67	7	used	use	VBN
cord-020871-1v6dcmt3	67	8	,	,	,
cord-020871-1v6dcmt3	67	9	we	-PRON-	PRP
cord-020871-1v6dcmt3	67	10	took	take	VBD
cord-020871-1v6dcmt3	67	11	all	all	DT
cord-020871-1v6dcmt3	67	12	standard	standard	JJ
cord-020871-1v6dcmt3	67	13	parameters	parameter	NNS
cord-020871-1v6dcmt3	67	14	(	(	-LRB-
cord-020871-1v6dcmt3	67	15	see	see	VB
cord-020871-1v6dcmt3	67	16	provided	provide	VBN
cord-020871-1v6dcmt3	67	17	code	code	NN
cord-020871-1v6dcmt3	67	18	6	6	CD
cord-020871-1v6dcmt3	67	19	)	)	-RRB-
cord-020871-1v6dcmt3	67	20	.	.	.
cord-020871-1v6dcmt3	68	1	Gensim	Gensim	NNP
cord-020871-1v6dcmt3	68	2	however	however	RB
cord-020871-1v6dcmt3	68	3	does	do	VBZ
cord-020871-1v6dcmt3	68	4	not	not	RB
cord-020871-1v6dcmt3	68	5	come	come	VB
cord-020871-1v6dcmt3	68	6	with	with	IN
cord-020871-1v6dcmt3	68	7	a	a	DT
cord-020871-1v6dcmt3	68	8	pre	pre	JJ
cord-020871-1v6dcmt3	68	9	-	-	JJ
cord-020871-1v6dcmt3	68	10	defined	define	VBN
cord-020871-1v6dcmt3	68	11	stopword	stopword	NN
cord-020871-1v6dcmt3	68	12	list	list	NN
cord-020871-1v6dcmt3	68	13	,	,	,
cord-020871-1v6dcmt3	68	14	and	and	CC
cord-020871-1v6dcmt3	68	15	therefore	therefore	RB
cord-020871-1v6dcmt3	68	16	,	,	,
cord-020871-1v6dcmt3	68	17	based	base	VBN
cord-020871-1v6dcmt3	68	18	on	on	IN
cord-020871-1v6dcmt3	68	19	our	-PRON-	PRP$
cord-020871-1v6dcmt3	68	20	own	own	JJ
cord-020871-1v6dcmt3	68	21	experience	experience	NN
cord-020871-1v6dcmt3	68	22	,	,	,
cord-020871-1v6dcmt3	68	23	we	-PRON-	PRP
cord-020871-1v6dcmt3	68	24	used	use	VBD
cord-020871-1v6dcmt3	68	25	the	the	DT
cord-020871-1v6dcmt3	68	26	one	one	NN
cord-020871-1v6dcmt3	68	27	provided	provide	VBN
cord-020871-1v6dcmt3	68	28	in	in	IN
cord-020871-1v6dcmt3	68	29	the	the	DT
cord-020871-1v6dcmt3	68	30	NLTK	NLTK	NNP
cord-020871-1v6dcmt3	68	31	library	library	NN
cord-020871-1v6dcmt3	68	32	7	7	CD
cord-020871-1v6dcmt3	68	33	for	for	IN
cord-020871-1v6dcmt3	68	34	English	English	NNP
cord-020871-1v6dcmt3	68	35	.	.	.
cord-020871-1v6dcmt3	69	1	For	for	IN
cord-020871-1v6dcmt3	69	2	the	the	DT
cord-020871-1v6dcmt3	69	3	second	second	JJ
cord-020871-1v6dcmt3	69	4	phase	phase	NN
cord-020871-1v6dcmt3	69	5	,	,	,
cord-020871-1v6dcmt3	69	6	transforming	transform	VBG
cord-020871-1v6dcmt3	69	7	terms	term	NNS
cord-020871-1v6dcmt3	69	8	and	and	CC
cord-020871-1v6dcmt3	69	9	documents	document	NNS
cord-020871-1v6dcmt3	69	10	to	to	IN
cord-020871-1v6dcmt3	69	11	vectors	vector	NNS
cord-020871-1v6dcmt3	69	12	,	,	,
cord-020871-1v6dcmt3	69	13	Zhang	Zhang	NNP
cord-020871-1v6dcmt3	69	14	et	et	NNP
cord-020871-1v6dcmt3	69	15	al	al	NNP
cord-020871-1v6dcmt3	69	16	.	.	.
cord-020871-1v6dcmt3	70	1	[	[	-LRB-
cord-020871-1v6dcmt3	70	2	14	14	CD
cord-020871-1v6dcmt3	70	3	]	]	-RRB-
cord-020871-1v6dcmt3	70	4	specify	specify	VBP
cord-020871-1v6dcmt3	70	5	that	that	IN
cord-020871-1v6dcmt3	70	6	all	all	DT
cord-020871-1v6dcmt3	70	7	trained	train	VBN
cord-020871-1v6dcmt3	70	8	models	model	NNS
cord-020871-1v6dcmt3	70	9	are	be	VBP
cord-020871-1v6dcmt3	70	10	50	50	CD
cord-020871-1v6dcmt3	70	11	dimensional	dimensional	JJ
cord-020871-1v6dcmt3	70	12	.	.	.
cord-020871-1v6dcmt3	71	1	We	-PRON-	PRP
cord-020871-1v6dcmt3	71	2	have	have	VBP
cord-020871-1v6dcmt3	71	3	additionally	additionally	RB
cord-020871-1v6dcmt3	71	4	experimented	experiment	VBN
cord-020871-1v6dcmt3	71	5	with	with	IN
cord-020871-1v6dcmt3	71	6	dimensionality	dimensionality	NN
cord-020871-1v6dcmt3	71	7	20	20	CD
cord-020871-1v6dcmt3	71	8	(	(	-LRB-
cord-020871-1v6dcmt3	71	9	used	use	VBN
cord-020871-1v6dcmt3	71	10	by	by	IN
cord-020871-1v6dcmt3	71	11	Clinchant	Clinchant	NNP
cord-020871-1v6dcmt3	71	12	and	and	CC
cord-020871-1v6dcmt3	71	13	Perronnin	Perronnin	NNP
cord-020871-1v6dcmt3	71	14	[	[	-LRB-
cord-020871-1v6dcmt3	71	15	3	3	CD
cord-020871-1v6dcmt3	71	16	]	]	-RRB-
cord-020871-1v6dcmt3	71	17	for	for	IN
cord-020871-1v6dcmt3	71	18	clustering	cluster	VBG
cord-020871-1v6dcmt3	71	19	)	)	-RRB-
cord-020871-1v6dcmt3	71	20	and	and	CC
cord-020871-1v6dcmt3	71	21	100	100	CD
cord-020871-1v6dcmt3	71	22	,	,	,
cord-020871-1v6dcmt3	71	23	as	as	IN
cord-020871-1v6dcmt3	71	24	we	-PRON-	PRP
cord-020871-1v6dcmt3	71	25	hypothesized	hypothesize	VBD
cord-020871-1v6dcmt3	71	26	that	that	IN
cord-020871-1v6dcmt3	71	27	50	50	CD
cord-020871-1v6dcmt3	71	28	might	may	MD
cord-020871-1v6dcmt3	71	29	be	be	VB
cord-020871-1v6dcmt3	71	30	too	too	RB
cord-020871-1v6dcmt3	71	31	low	low	JJ
cord-020871-1v6dcmt3	71	32	.	.	.
cord-020871-1v6dcmt3	72	1	The	the	DT
cord-020871-1v6dcmt3	72	2	TF	TF	NNP
cord-020871-1v6dcmt3	72	3	-	-	HYPH
cord-020871-1v6dcmt3	72	4	IDF	IDF	NNP
cord-020871-1v6dcmt3	72	5	model	model	NN
cord-020871-1v6dcmt3	72	6	is	be	VBZ
cord-020871-1v6dcmt3	72	7	5000	5000	CD
cord-020871-1v6dcmt3	72	8	dimensional	dimensional	JJ
cord-020871-1v6dcmt3	72	9	(	(	-LRB-
cord-020871-1v6dcmt3	72	10	i.e.	i.e.	FW
cord-020871-1v6dcmt3	72	11	only	only	RB
cord-020871-1v6dcmt3	72	12	the	the	DT
cord-020871-1v6dcmt3	72	13	top	top	JJ
cord-020871-1v6dcmt3	72	14	5000	5000	CD
cord-020871-1v6dcmt3	72	15	terms	term	NNS
cord-020871-1v6dcmt3	72	16	based	base	VBN
cord-020871-1v6dcmt3	72	17	on	on	IN
cord-020871-1v6dcmt3	72	18	their	-PRON-	PRP$
cord-020871-1v6dcmt3	72	19	tf	tf	NNP
cord-020871-1v6dcmt3	72	20	-	-	HYPH
cord-020871-1v6dcmt3	72	21	idf	idf	NNP
cord-020871-1v6dcmt3	72	22	value	value	NN
cord-020871-1v6dcmt3	72	23	are	be	VBP
cord-020871-1v6dcmt3	72	24	used	use	VBN
cord-020871-1v6dcmt3	72	25	)	)	-RRB-
cord-020871-1v6dcmt3	72	26	,	,	,
cord-020871-1v6dcmt3	72	27	while	while	IN
cord-020871-1v6dcmt3	72	28	the	the	DT
cord-020871-1v6dcmt3	72	29	Fischer	Fischer	NNP
cord-020871-1v6dcmt3	72	30	-	-	HYPH
cord-020871-1v6dcmt3	72	31	Kernel	Kernel	NNP
cord-020871-1v6dcmt3	72	32	models	model	NNS
cord-020871-1v6dcmt3	72	33	are	be	VBP
cord-020871-1v6dcmt3	72	34	15	15	CD
cord-020871-1v6dcmt3	72	35	×	×	NN
cord-020871-1v6dcmt3	72	36	d	d	NN
cord-020871-1v6dcmt3	72	37	dimensional	dimensional	JJ
cord-020871-1v6dcmt3	72	38	,	,	,
cord-020871-1v6dcmt3	72	39	where	where	WRB
cord-020871-1v6dcmt3	72	40	d	d	NN
cord-020871-1v6dcmt3	72	41	=	=	NN
cord-020871-1v6dcmt3	72	42	{	{	-LRB-
cord-020871-1v6dcmt3	72	43	20	20	CD
cord-020871-1v6dcmt3	72	44	,	,	,
cord-020871-1v6dcmt3	72	45	50	50	CD
cord-020871-1v6dcmt3	72	46	,	,	,
cord-020871-1v6dcmt3	72	47	100	100	CD
cord-020871-1v6dcmt3	72	48	}	}	-RRB-
cord-020871-1v6dcmt3	72	49	,	,	,
cord-020871-1v6dcmt3	72	50	as	as	IN
cord-020871-1v6dcmt3	72	51	just	just	RB
cord-020871-1v6dcmt3	72	52	explained	explain	VBN
cord-020871-1v6dcmt3	72	53	.	.	.
cord-020871-1v6dcmt3	73	1	In	in	IN
cord-020871-1v6dcmt3	73	2	what	what	WP
cord-020871-1v6dcmt3	73	3	follows	follow	VBZ
cord-020871-1v6dcmt3	73	4	,	,	,
cord-020871-1v6dcmt3	73	5	d	d	NNP
cord-020871-1v6dcmt3	73	6	refers	refer	VBZ
cord-020871-1v6dcmt3	73	7	to	to	IN
cord-020871-1v6dcmt3	73	8	the	the	DT
cord-020871-1v6dcmt3	73	9	dimensionality	dimensionality	NN
cord-020871-1v6dcmt3	73	10	of	of	IN
cord-020871-1v6dcmt3	73	11	LSI	LSI	NNP
cord-020871-1v6dcmt3	73	12	,	,	,
cord-020871-1v6dcmt3	73	13	LDA	LDA	NNP
cord-020871-1v6dcmt3	73	14	,	,	,
cord-020871-1v6dcmt3	73	15	cBow	cBow	NNP
cord-020871-1v6dcmt3	73	16	,	,	,
cord-020871-1v6dcmt3	73	17	and	and	CC
cord-020871-1v6dcmt3	73	18	PV	pv	NN
cord-020871-1v6dcmt3	73	19	models	model	NNS
cord-020871-1v6dcmt3	73	20	.	.	.
cord-020871-1v6dcmt3	74	1	The	the	DT
cord-020871-1v6dcmt3	74	2	cBoW	cbow	NN
cord-020871-1v6dcmt3	74	3	and	and	CC
cord-020871-1v6dcmt3	74	4	PV	pv	NN
cord-020871-1v6dcmt3	74	5	models	model	NNS
cord-020871-1v6dcmt3	74	6	are	be	VBP
cord-020871-1v6dcmt3	74	7	trained	train	VBN
cord-020871-1v6dcmt3	74	8	using	use	VBG
cord-020871-1v6dcmt3	74	9	a	a	DT
cord-020871-1v6dcmt3	74	10	default	default	NN
cord-020871-1v6dcmt3	74	11	window	window	NN
cord-020871-1v6dcmt3	74	12	size	size	NN
cord-020871-1v6dcmt3	74	13	of	of	IN
cord-020871-1v6dcmt3	74	14	5	5	CD
cord-020871-1v6dcmt3	74	15	,	,	,
cord-020871-1v6dcmt3	74	16	keeping	keep	VBG
cord-020871-1v6dcmt3	74	17	both	both	CC
cord-020871-1v6dcmt3	74	18	low	low	JJ
cord-020871-1v6dcmt3	74	19	and	and	CC
cord-020871-1v6dcmt3	74	20	high	high	JJ
cord-020871-1v6dcmt3	74	21	-	-	HYPH
cord-020871-1v6dcmt3	74	22	frequency	frequency	NN
cord-020871-1v6dcmt3	74	23	terms	term	NNS
cord-020871-1v6dcmt3	74	24	,	,	,
cord-020871-1v6dcmt3	74	25	again	again	RB
cord-020871-1v6dcmt3	74	26	following	follow	VBG
cord-020871-1v6dcmt3	74	27	the	the	DT
cord-020871-1v6dcmt3	74	28	setup	setup	NN
cord-020871-1v6dcmt3	74	29	of	of	IN
cord-020871-1v6dcmt3	74	30	the	the	DT
cord-020871-1v6dcmt3	74	31	original	original	JJ
cord-020871-1v6dcmt3	74	32	experiment	experiment	NN
cord-020871-1v6dcmt3	74	33	.	.	.
cord-020871-1v6dcmt3	75	1	The	the	DT
cord-020871-1v6dcmt3	75	2	LDA	LDA	NNP
cord-020871-1v6dcmt3	75	3	model	model	NN
cord-020871-1v6dcmt3	75	4	is	be	VBZ
cord-020871-1v6dcmt3	75	5	trained	train	VBN
cord-020871-1v6dcmt3	75	6	using	use	VBG
cord-020871-1v6dcmt3	75	7	a	a	DT
cord-020871-1v6dcmt3	75	8	chunk	chunk	NN
cord-020871-1v6dcmt3	75	9	size	size	NN
cord-020871-1v6dcmt3	75	10	of	of	IN
cord-020871-1v6dcmt3	75	11	1000	1000	CD
cord-020871-1v6dcmt3	75	12	documents	document	NNS
cord-020871-1v6dcmt3	75	13	and	and	CC
cord-020871-1v6dcmt3	75	14	for	for	IN
cord-020871-1v6dcmt3	75	15	a	a	DT
cord-020871-1v6dcmt3	75	16	number	number	NN
cord-020871-1v6dcmt3	75	17	of	of	IN
cord-020871-1v6dcmt3	75	18	iterations	iteration	NNS
cord-020871-1v6dcmt3	75	19	over	over	IN
cord-020871-1v6dcmt3	75	20	the	the	DT
cord-020871-1v6dcmt3	75	21	corpus	corpus	NN
cord-020871-1v6dcmt3	75	22	ranging	range	VBG
cord-020871-1v6dcmt3	75	23	from	from	IN
cord-020871-1v6dcmt3	75	24	20	20	CD
cord-020871-1v6dcmt3	75	25	to	to	TO
cord-020871-1v6dcmt3	75	26	100	100	CD
cord-020871-1v6dcmt3	75	27	.	.	.
cord-020871-1v6dcmt3	76	1	For	for	IN
cord-020871-1v6dcmt3	76	2	the	the	DT
cord-020871-1v6dcmt3	76	3	FK	FK	NNP
cord-020871-1v6dcmt3	76	4	methods	method	NNS
cord-020871-1v6dcmt3	76	5	,	,	,
cord-020871-1v6dcmt3	76	6	both	both	DT
cord-020871-1v6dcmt3	76	7	fitting	fitting	JJ
cord-020871-1v6dcmt3	76	8	procedures	procedure	NNS
cord-020871-1v6dcmt3	76	9	(	(	-LRB-
cord-020871-1v6dcmt3	76	10	GMM	GMM	NNP
cord-020871-1v6dcmt3	76	11	and	and	CC
cord-020871-1v6dcmt3	76	12	moVMF	movmf	NN
cord-020871-1v6dcmt3	76	13	)	)	-RRB-
cord-020871-1v6dcmt3	76	14	are	be	VBP
cord-020871-1v6dcmt3	76	15	independently	independently	RB
cord-020871-1v6dcmt3	76	16	initialised	initialise	VBN
cord-020871-1v6dcmt3	76	17	10	10	CD
cord-020871-1v6dcmt3	76	18	times	time	NNS
cord-020871-1v6dcmt3	76	19	and	and	CC
cord-020871-1v6dcmt3	76	20	the	the	DT
cord-020871-1v6dcmt3	76	21	best	good	JJS
cord-020871-1v6dcmt3	76	22	fitting	fitting	JJ
cord-020871-1v6dcmt3	76	23	model	model	NN
cord-020871-1v6dcmt3	76	24	is	be	VBZ
cord-020871-1v6dcmt3	76	25	kept	keep	VBN
cord-020871-1v6dcmt3	76	26	.	.	.
cord-020871-1v6dcmt3	77	1	For	for	IN
cord-020871-1v6dcmt3	77	2	the	the	DT
cord-020871-1v6dcmt3	77	3	third	third	JJ
cord-020871-1v6dcmt3	77	4	phase	phase	NN
cord-020871-1v6dcmt3	77	5	,	,	,
cord-020871-1v6dcmt3	77	6	parameters	parameter	NNS
cord-020871-1v6dcmt3	77	7	are	be	VBP
cord-020871-1v6dcmt3	77	8	explained	explain	VBN
cord-020871-1v6dcmt3	77	9	in	in	IN
cord-020871-1v6dcmt3	77	10	the	the	DT
cord-020871-1v6dcmt3	77	11	following	follow	VBG
cord-020871-1v6dcmt3	77	12	sections	section	NNS
cord-020871-1v6dcmt3	77	13	.	.	.
cord-020871-1v6dcmt3	78	1	Logistic	logistic	JJ
cord-020871-1v6dcmt3	78	2	regression	regression	NN
cord-020871-1v6dcmt3	78	3	is	be	VBZ
cord-020871-1v6dcmt3	78	4	used	use	VBN
cord-020871-1v6dcmt3	78	5	for	for	IN
cord-020871-1v6dcmt3	78	6	classification	classification	NN
cord-020871-1v6dcmt3	78	7	in	in	IN
cord-020871-1v6dcmt3	78	8	Zhang	Zhang	NNP
cord-020871-1v6dcmt3	78	9	et	et	NNP
cord-020871-1v6dcmt3	78	10	al	al	NNP
cord-020871-1v6dcmt3	78	11	.	.	NNP
cord-020871-1v6dcmt3	78	12	,	,	,
cord-020871-1v6dcmt3	78	13	and	and	CC
cord-020871-1v6dcmt3	78	14	therefore	therefore	RB
cord-020871-1v6dcmt3	78	15	also	also	RB
cord-020871-1v6dcmt3	78	16	used	use	VBD
cord-020871-1v6dcmt3	78	17	here	here	RB
cord-020871-1v6dcmt3	78	18	.	.	.
cord-020871-1v6dcmt3	79	1	The	the	DT
cord-020871-1v6dcmt3	79	2	results	result	NNS
cord-020871-1v6dcmt3	79	3	of	of	IN
cord-020871-1v6dcmt3	79	4	our	-PRON-	PRP$
cord-020871-1v6dcmt3	79	5	experiments	experiment	NNS
cord-020871-1v6dcmt3	79	6	,	,	,
cord-020871-1v6dcmt3	79	7	for	for	IN
cord-020871-1v6dcmt3	79	8	d	d	NN
cord-020871-1v6dcmt3	79	9	=	=	SYM
cord-020871-1v6dcmt3	79	10	50	50	CD
cord-020871-1v6dcmt3	79	11	and	and	CC
cord-020871-1v6dcmt3	79	12	100-dimensional	100-dimensional	CD
cord-020871-1v6dcmt3	79	13	feature	feature	NN
cord-020871-1v6dcmt3	79	14	vectors	vector	NNS
cord-020871-1v6dcmt3	79	15	,	,	,
cord-020871-1v6dcmt3	79	16	are	be	VBP
cord-020871-1v6dcmt3	79	17	summarised	summarise	VBN
cord-020871-1v6dcmt3	79	18	in	in	IN
cord-020871-1v6dcmt3	79	19	Table	Table	NNP
cord-020871-1v6dcmt3	79	20	1	1	CD
cord-020871-1v6dcmt3	79	21	.	.	.
cord-020871-1v6dcmt3	80	1	For	for	IN
cord-020871-1v6dcmt3	80	2	all	all	PDT
cord-020871-1v6dcmt3	80	3	the	the	DT
cord-020871-1v6dcmt3	80	4	methods	method	NNS
cord-020871-1v6dcmt3	80	5	,	,	,
cord-020871-1v6dcmt3	80	6	we	-PRON-	PRP
cord-020871-1v6dcmt3	80	7	perform	perform	VBP
cord-020871-1v6dcmt3	80	8	a	a	DT
cord-020871-1v6dcmt3	80	9	parameter	parameter	NN
cord-020871-1v6dcmt3	80	10	scan	scan	NN
cord-020871-1v6dcmt3	80	11	of	of	IN
cord-020871-1v6dcmt3	80	12	the	the	DT
cord-020871-1v6dcmt3	80	13	(	(	-LRB-
cord-020871-1v6dcmt3	80	14	inverse	inverse	NN
cord-020871-1v6dcmt3	80	15	)	)	-RRB-
cord-020871-1v6dcmt3	80	16	regularisation	regularisation	NN
cord-020871-1v6dcmt3	80	17	strength	strength	NN
cord-020871-1v6dcmt3	80	18	of	of	IN
cord-020871-1v6dcmt3	80	19	the	the	DT
cord-020871-1v6dcmt3	80	20	logistic	logistic	JJ
cord-020871-1v6dcmt3	80	21	regression	regression	NN
cord-020871-1v6dcmt3	80	22	classifier	classifier	NN
cord-020871-1v6dcmt3	80	23	,	,	,
cord-020871-1v6dcmt3	80	24	as	as	IN
cord-020871-1v6dcmt3	80	25	shown	show	VBN
cord-020871-1v6dcmt3	80	26	in	in	IN
cord-020871-1v6dcmt3	80	27	Fig	Fig	NNP
cord-020871-1v6dcmt3	80	28	.	.	NNP
cord-020871-1v6dcmt3	80	29	1(a	1(a	CD
cord-020871-1v6dcmt3	80	30	)	)	-RRB-
cord-020871-1v6dcmt3	81	1	and	and	CC
cord-020871-1v6dcmt3	81	2	(	(	-LRB-
cord-020871-1v6dcmt3	81	3	b	b	NN
cord-020871-1v6dcmt3	81	4	)	)	-RRB-
cord-020871-1v6dcmt3	81	5	.	.	.
cord-020871-1v6dcmt3	82	1	Additionally	additionally	RB
cord-020871-1v6dcmt3	82	2	,	,	,
cord-020871-1v6dcmt3	82	3	the	the	DT
cord-020871-1v6dcmt3	82	4	learning	learn	VBG
cord-020871-1v6dcmt3	82	5	algorithms	algorithm	NNS
cord-020871-1v6dcmt3	82	6	are	be	VBP
cord-020871-1v6dcmt3	82	7	trained	train	VBN
cord-020871-1v6dcmt3	82	8	for	for	IN
cord-020871-1v6dcmt3	82	9	a	a	DT
cord-020871-1v6dcmt3	82	10	different	different	JJ
cord-020871-1v6dcmt3	82	11	number	number	NN
cord-020871-1v6dcmt3	82	12	of	of	IN
cord-020871-1v6dcmt3	82	13	epochs	epoch	NNS
cord-020871-1v6dcmt3	82	14	and	and	CC
cord-020871-1v6dcmt3	82	15	the	the	DT
cord-020871-1v6dcmt3	82	16	resulting	result	VBG
cord-020871-1v6dcmt3	82	17	classification	classification	NN
cord-020871-1v6dcmt3	82	18	accuracy	accuracy	NN
cord-020871-1v6dcmt3	82	19	assessed	assess	VBN
cord-020871-1v6dcmt3	82	20	,	,	,
cord-020871-1v6dcmt3	82	21	cf	cf	NNP
cord-020871-1v6dcmt3	82	22	.	.	.
cord-020871-1v6dcmt3	83	1	Fig	Fig	NNP
cord-020871-1v6dcmt3	83	2	.	.	.
cord-020871-1v6dcmt3	83	3	1(c	1(c	CD
cord-020871-1v6dcmt3	83	4	)	)	-RRB-
cord-020871-1v6dcmt3	83	5	and	and	CC
cord-020871-1v6dcmt3	83	6	(	(	-LRB-
cord-020871-1v6dcmt3	83	7	d	d	NN
cord-020871-1v6dcmt3	83	8	)	)	-RRB-
cord-020871-1v6dcmt3	83	9	.	.	.
cord-020871-1v6dcmt3	84	1	Figure	figure	NN
cord-020871-1v6dcmt3	84	2	1	1	CD
cord-020871-1v6dcmt3	84	3	(	(	-LRB-
cord-020871-1v6dcmt3	84	4	a	a	NN
cord-020871-1v6dcmt3	84	5	)	)	-RRB-
cord-020871-1v6dcmt3	84	6	indicates	indicate	VBZ
cord-020871-1v6dcmt3	84	7	that	that	IN
cord-020871-1v6dcmt3	84	8	cBow	cBow	NNP
cord-020871-1v6dcmt3	84	9	,	,	,
cord-020871-1v6dcmt3	84	10	FV	FV	NNP
cord-020871-1v6dcmt3	84	11	-	-	HYPH
cord-020871-1v6dcmt3	84	12	GMM	GMM	NNP
cord-020871-1v6dcmt3	84	13	,	,	,
cord-020871-1v6dcmt3	84	14	FV	FV	NNP
cord-020871-1v6dcmt3	84	15	-	-	HYPH
cord-020871-1v6dcmt3	84	16	moVMF	movmf	NN
cord-020871-1v6dcmt3	84	17	,	,	,
cord-020871-1v6dcmt3	84	18	and	and	CC
cord-020871-1v6dcmt3	84	19	the	the	DT
cord-020871-1v6dcmt3	84	20	simple	simple	JJ
cord-020871-1v6dcmt3	84	21	TF	TF	NNP
cord-020871-1v6dcmt3	84	22	-	-	HYPH
cord-020871-1v6dcmt3	84	23	IDF	IDF	NNP
cord-020871-1v6dcmt3	84	24	,	,	,
cord-020871-1v6dcmt3	84	25	when	when	WRB
cord-020871-1v6dcmt3	84	26	properly	properly	RB
cord-020871-1v6dcmt3	84	27	tuned	tune	VBN
cord-020871-1v6dcmt3	84	28	,	,	,
cord-020871-1v6dcmt3	84	29	exhibit	exhibit	VBP
cord-020871-1v6dcmt3	84	30	a	a	DT
cord-020871-1v6dcmt3	84	31	very	very	RB
cord-020871-1v6dcmt3	84	32	similar	similar	JJ
cord-020871-1v6dcmt3	84	33	accuracy	accuracy	NN
cord-020871-1v6dcmt3	84	34	on	on	IN
cord-020871-1v6dcmt3	84	35	subj	subj	NNP
cord-020871-1v6dcmt3	85	1	-the	-the	DT
cord-020871-1v6dcmt3	85	2	given	give	VBN
cord-020871-1v6dcmt3	85	3	confidence	confidence	NN
cord-020871-1v6dcmt3	85	4	intervals	interval	NNS
cord-020871-1v6dcmt3	85	5	do	do	VBP
cord-020871-1v6dcmt3	85	6	not	not	RB
cord-020871-1v6dcmt3	85	7	indeed	indeed	RB
cord-020871-1v6dcmt3	85	8	allow	allow	VB
cord-020871-1v6dcmt3	85	9	us	-PRON-	PRP
cord-020871-1v6dcmt3	85	10	to	to	TO
cord-020871-1v6dcmt3	85	11	identify	identify	VB
cord-020871-1v6dcmt3	85	12	a	a	DT
cord-020871-1v6dcmt3	85	13	single	single	JJ
cord-020871-1v6dcmt3	85	14	,	,	,
cord-020871-1v6dcmt3	85	15	best	good	JJS
cord-020871-1v6dcmt3	85	16	model	model	NN
cord-020871-1v6dcmt3	85	17	.	.	.
cord-020871-1v6dcmt3	86	1	Surprisingly	surprisingly	RB
cord-020871-1v6dcmt3	86	2	,	,	,
cord-020871-1v6dcmt3	86	3	TF	TF	NNP
cord-020871-1v6dcmt3	86	4	-	-	HYPH
cord-020871-1v6dcmt3	86	5	IDF	IDF	NNP
cord-020871-1v6dcmt3	86	6	outperforms	outperform	VBZ
cord-020871-1v6dcmt3	86	7	all	all	PDT
cord-020871-1v6dcmt3	86	8	the	the	DT
cord-020871-1v6dcmt3	86	9	others	other	NNS
cord-020871-1v6dcmt3	86	10	on	on	IN
cord-020871-1v6dcmt3	86	11	the	the	DT
cord-020871-1v6dcmt3	86	12	sent	send	VBN
cord-020871-1v6dcmt3	86	13	dataset	dataset	NN
cord-020871-1v6dcmt3	86	14	(	(	-LRB-
cord-020871-1v6dcmt3	86	15	Fig	Fig	NNP
cord-020871-1v6dcmt3	86	16	.	.	.
cord-020871-1v6dcmt3	86	17	1(b	1(b	CD
cord-020871-1v6dcmt3	86	18	)	)	-RRB-
cord-020871-1v6dcmt3	86	19	)	)	-RRB-
cord-020871-1v6dcmt3	86	20	.	.	.
cord-020871-1v6dcmt3	87	1	Increasing	increase	VBG
cord-020871-1v6dcmt3	87	2	the	the	DT
cord-020871-1v6dcmt3	87	3	dimensionality	dimensionality	NN
cord-020871-1v6dcmt3	87	4	of	of	IN
cord-020871-1v6dcmt3	87	5	the	the	DT
cord-020871-1v6dcmt3	87	6	feature	feature	NN
cord-020871-1v6dcmt3	87	7	vectors	vector	NNS
cord-020871-1v6dcmt3	87	8	,	,	,
cord-020871-1v6dcmt3	87	9	from	from	IN
cord-020871-1v6dcmt3	87	10	d	d	NN
cord-020871-1v6dcmt3	87	11	=	=	SYM
cord-020871-1v6dcmt3	87	12	50	50	CD
cord-020871-1v6dcmt3	87	13	to	to	TO
cord-020871-1v6dcmt3	87	14	100	100	CD
cord-020871-1v6dcmt3	87	15	,	,	,
cord-020871-1v6dcmt3	87	16	has	have	VBZ
cord-020871-1v6dcmt3	87	17	the	the	DT
cord-020871-1v6dcmt3	87	18	effect	effect	NN
cord-020871-1v6dcmt3	87	19	of	of	IN
cord-020871-1v6dcmt3	87	20	reducing	reduce	VBG
cord-020871-1v6dcmt3	87	21	the	the	DT
cord-020871-1v6dcmt3	87	22	gap	gap	NN
cord-020871-1v6dcmt3	87	23	between	between	IN
cord-020871-1v6dcmt3	87	24	TF	TF	NNP
cord-020871-1v6dcmt3	87	25	-	-	HYPH
cord-020871-1v6dcmt3	87	26	IDF	IDF	NNP
cord-020871-1v6dcmt3	87	27	and	and	CC
cord-020871-1v6dcmt3	87	28	the	the	DT
cord-020871-1v6dcmt3	87	29	rest	rest	NN
cord-020871-1v6dcmt3	87	30	of	of	IN
cord-020871-1v6dcmt3	87	31	the	the	DT
cord-020871-1v6dcmt3	87	32	models	model	NNS
cord-020871-1v6dcmt3	87	33	on	on	IN
cord-020871-1v6dcmt3	87	34	the	the	DT
cord-020871-1v6dcmt3	87	35	sent	send	VBN
cord-020871-1v6dcmt3	87	36	dataset	dataset	NN
cord-020871-1v6dcmt3	87	37	(	(	-LRB-
cord-020871-1v6dcmt3	87	38	see	see	VB
cord-020871-1v6dcmt3	87	39	Table	table	NN
cord-020871-1v6dcmt3	87	40	1	1	CD
cord-020871-1v6dcmt3	87	41	)	)	-RRB-
cord-020871-1v6dcmt3	87	42	.	.	.
cord-020871-1v6dcmt3	88	1	For	for	IN
cord-020871-1v6dcmt3	88	2	clustering	cluster	VBG
cord-020871-1v6dcmt3	88	3	experiments	experiment	NNS
cord-020871-1v6dcmt3	88	4	,	,	,
cord-020871-1v6dcmt3	88	5	the	the	DT
cord-020871-1v6dcmt3	88	6	obtained	obtain	VBN
cord-020871-1v6dcmt3	88	7	feature	feature	NN
cord-020871-1v6dcmt3	88	8	vectors	vector	NNS
cord-020871-1v6dcmt3	88	9	are	be	VBP
cord-020871-1v6dcmt3	88	10	passed	pass	VBN
cord-020871-1v6dcmt3	88	11	to	to	IN
cord-020871-1v6dcmt3	88	12	the	the	DT
cord-020871-1v6dcmt3	88	13	kmeans	kmeans	NNP
cord-020871-1v6dcmt3	88	14	algorithm	algorithm	NN
cord-020871-1v6dcmt3	88	15	.	.	.
cord-020871-1v6dcmt3	89	1	The	the	DT
cord-020871-1v6dcmt3	89	2	results	result	NNS
cord-020871-1v6dcmt3	89	3	of	of	IN
cord-020871-1v6dcmt3	89	4	our	-PRON-	PRP$
cord-020871-1v6dcmt3	89	5	experiments	experiment	NNS
cord-020871-1v6dcmt3	89	6	,	,	,
cord-020871-1v6dcmt3	89	7	measured	measure	VBN
cord-020871-1v6dcmt3	89	8	in	in	IN
cord-020871-1v6dcmt3	89	9	terms	term	NNS
cord-020871-1v6dcmt3	89	10	of	of	IN
cord-020871-1v6dcmt3	89	11	Adjusted	Adjusted	NNP
cord-020871-1v6dcmt3	89	12	Rand	Rand	NNP
cord-020871-1v6dcmt3	89	13	Index	Index	NNP
cord-020871-1v6dcmt3	89	14	(	(	-LRB-
cord-020871-1v6dcmt3	89	15	ARI	ARI	NNP
cord-020871-1v6dcmt3	89	16	)	)	-RRB-
cord-020871-1v6dcmt3	89	17	and	and	CC
cord-020871-1v6dcmt3	89	18	Normalized	Normalized	NNP
cord-020871-1v6dcmt3	89	19	Mutual	Mutual	NNP
cord-020871-1v6dcmt3	89	20	Information	Information	NNP
cord-020871-1v6dcmt3	89	21	(	(	-LRB-
cord-020871-1v6dcmt3	89	22	NMI	NMI	NNP
cord-020871-1v6dcmt3	89	23	)	)	-RRB-
cord-020871-1v6dcmt3	89	24	,	,	,
cord-020871-1v6dcmt3	89	25	are	be	VBP
cord-020871-1v6dcmt3	89	26	summarised	summarise	VBN
cord-020871-1v6dcmt3	89	27	in	in	IN
cord-020871-1v6dcmt3	89	28	Table	Table	NNP
cord-020871-1v6dcmt3	89	29	2	2	CD
cord-020871-1v6dcmt3	89	30	.	.	.
cord-020871-1v6dcmt3	90	1	We	-PRON-	PRP
cord-020871-1v6dcmt3	90	2	used	use	VBD
cord-020871-1v6dcmt3	90	3	both	both	CC
cord-020871-1v6dcmt3	90	4	d	d	NN
cord-020871-1v6dcmt3	90	5	=	=	SYM
cord-020871-1v6dcmt3	90	6	20	20	CD
cord-020871-1v6dcmt3	90	7	and	and	CC
cord-020871-1v6dcmt3	90	8	50-dimensional	50-dimensional	CD
cord-020871-1v6dcmt3	90	9	feature	feature	NN
cord-020871-1v6dcmt3	90	10	vectors	vector	NNS
cord-020871-1v6dcmt3	90	11	.	.	.
cord-020871-1v6dcmt3	91	1	Note	note	VB
cord-020871-1v6dcmt3	91	2	that	that	IN
cord-020871-1v6dcmt3	91	3	the	the	DT
cord-020871-1v6dcmt3	91	4	evaluation	evaluation	NN
cord-020871-1v6dcmt3	91	5	of	of	IN
cord-020871-1v6dcmt3	91	6	the	the	DT
cord-020871-1v6dcmt3	91	7	clustering	clustering	NN
cord-020871-1v6dcmt3	91	8	algorithms	algorithm	NNS
cord-020871-1v6dcmt3	91	9	is	be	VBZ
cord-020871-1v6dcmt3	91	10	based	base	VBN
cord-020871-1v6dcmt3	91	11	on	on	IN
cord-020871-1v6dcmt3	91	12	the	the	DT
cord-020871-1v6dcmt3	91	13	knowledge	knowledge	NN
cord-020871-1v6dcmt3	91	14	of	of	IN
cord-020871-1v6dcmt3	91	15	the	the	DT
cord-020871-1v6dcmt3	91	16	ground	ground	NN
cord-020871-1v6dcmt3	91	17	truth	truth	NN
cord-020871-1v6dcmt3	91	18	class	class	NN
cord-020871-1v6dcmt3	91	19	assignments	assignment	NNS
cord-020871-1v6dcmt3	91	20	,	,	,
cord-020871-1v6dcmt3	91	21	available	available	JJ
cord-020871-1v6dcmt3	91	22	in	in	IN
cord-020871-1v6dcmt3	91	23	the	the	DT
cord-020871-1v6dcmt3	91	24	20	20	CD
cord-020871-1v6dcmt3	91	25	Newsgroups	newsgroup	NNS
cord-020871-1v6dcmt3	91	26	dataset	dataset	NN
cord-020871-1v6dcmt3	91	27	.	.	.
cord-020871-1v6dcmt3	92	1	As	as	IN
cord-020871-1v6dcmt3	92	2	opposed	oppose	VBN
cord-020871-1v6dcmt3	92	3	to	to	IN
cord-020871-1v6dcmt3	92	4	classification	classification	NN
cord-020871-1v6dcmt3	92	5	,	,	,
cord-020871-1v6dcmt3	92	6	clustering	clustering	NN
cord-020871-1v6dcmt3	92	7	experiments	experiment	NNS
cord-020871-1v6dcmt3	92	8	show	show	VBP
cord-020871-1v6dcmt3	92	9	a	a	DT
cord-020871-1v6dcmt3	92	10	generous	generous	JJ
cord-020871-1v6dcmt3	92	11	imbalance	imbalance	NN
cord-020871-1v6dcmt3	92	12	in	in	IN
cord-020871-1v6dcmt3	92	13	performance	performance	NN
cord-020871-1v6dcmt3	92	14	and	and	CC
cord-020871-1v6dcmt3	92	15	firmly	firmly	RB
cord-020871-1v6dcmt3	92	16	speak	speak	VBP
cord-020871-1v6dcmt3	92	17	in	in	IN
cord-020871-1v6dcmt3	92	18	favour	favour	NN
cord-020871-1v6dcmt3	92	19	of	of	IN
cord-020871-1v6dcmt3	92	20	PV	pv	NN
cord-020871-1v6dcmt3	92	21	-	-	HYPH
cord-020871-1v6dcmt3	92	22	DBOW	DBOW	NNP
cord-020871-1v6dcmt3	92	23	.	.	.
cord-020871-1v6dcmt3	93	1	Interestingly	interestingly	RB
cord-020871-1v6dcmt3	93	2	,	,	,
cord-020871-1v6dcmt3	93	3	TF	TF	NNP
cord-020871-1v6dcmt3	93	4	-	-	HYPH
cord-020871-1v6dcmt3	93	5	IDF	IDF	NNP
cord-020871-1v6dcmt3	93	6	,	,	,
cord-020871-1v6dcmt3	93	7	FV	FV	NNP
cord-020871-1v6dcmt3	93	8	-	-	HYPH
cord-020871-1v6dcmt3	93	9	GMM	GMM	NNP
cord-020871-1v6dcmt3	93	10	,	,	,
cord-020871-1v6dcmt3	93	11	and	and	CC
cord-020871-1v6dcmt3	93	12	FV	fv	NN
cord-020871-1v6dcmt3	93	13	-	-	HYPH
cord-020871-1v6dcmt3	93	14	moVMF	movmf	NN
cord-020871-1v6dcmt3	93	15	,	,	,
cord-020871-1v6dcmt3	93	16	all	all	DT
cord-020871-1v6dcmt3	93	17	providing	provide	VBG
cord-020871-1v6dcmt3	93	18	high	high	JJ
cord-020871-1v6dcmt3	93	19	-	-	HYPH
cord-020871-1v6dcmt3	93	20	dimensional	dimensional	JJ
cord-020871-1v6dcmt3	93	21	document	document	NN
cord-020871-1v6dcmt3	93	22	representations	representation	NNS
cord-020871-1v6dcmt3	93	23	,	,	,
cord-020871-1v6dcmt3	93	24	have	have	VBP
cord-020871-1v6dcmt3	93	25	a	a	DT
cord-020871-1v6dcmt3	93	26	low	low	JJ
cord-020871-1v6dcmt3	93	27	clustering	clustering	NN
cord-020871-1v6dcmt3	93	28	effectiveness	effectiveness	NN
cord-020871-1v6dcmt3	93	29	.	.	.
cord-020871-1v6dcmt3	94	1	LSI	LSI	NNP
cord-020871-1v6dcmt3	94	2	and	and	CC
cord-020871-1v6dcmt3	94	3	LDA	LDA	NNP
cord-020871-1v6dcmt3	94	4	achieve	achieve	VB
cord-020871-1v6dcmt3	94	5	low	low	JJ
cord-020871-1v6dcmt3	94	6	accuracy	accuracy	NN
cord-020871-1v6dcmt3	94	7	(	(	-LRB-
cord-020871-1v6dcmt3	94	8	see	see	VB
cord-020871-1v6dcmt3	94	9	Table	table	NN
cord-020871-1v6dcmt3	94	10	1	1	CD
cord-020871-1v6dcmt3	94	11	)	)	-RRB-
cord-020871-1v6dcmt3	94	12	and	and	CC
cord-020871-1v6dcmt3	94	13	are	be	VBP
cord-020871-1v6dcmt3	94	14	omitted	omit	VBN
cord-020871-1v6dcmt3	94	15	here	here	RB
cord-020871-1v6dcmt3	94	16	for	for	IN
cord-020871-1v6dcmt3	94	17	visibility	visibility	NN
cord-020871-1v6dcmt3	94	18	.	.	.
cord-020871-1v6dcmt3	95	1	The	the	DT
cord-020871-1v6dcmt3	95	2	left	left	JJ
cord-020871-1v6dcmt3	95	3	panels	panel	NNS
cord-020871-1v6dcmt3	95	4	[	[	-LRB-
cord-020871-1v6dcmt3	95	5	(	(	-LRB-
cord-020871-1v6dcmt3	95	6	a	a	LS
cord-020871-1v6dcmt3	95	7	)	)	-RRB-
cord-020871-1v6dcmt3	95	8	and	and	CC
cord-020871-1v6dcmt3	95	9	(	(	-LRB-
cord-020871-1v6dcmt3	95	10	b	b	NN
cord-020871-1v6dcmt3	95	11	)	)	-RRB-
cord-020871-1v6dcmt3	95	12	]	]	-RRB-
cord-020871-1v6dcmt3	95	13	show	show	VBP
cord-020871-1v6dcmt3	95	14	the	the	DT
cord-020871-1v6dcmt3	95	15	effect	effect	NN
cord-020871-1v6dcmt3	95	16	of	of	IN
cord-020871-1v6dcmt3	95	17	(	(	-LRB-
cord-020871-1v6dcmt3	95	18	inverse	inverse	NN
cord-020871-1v6dcmt3	95	19	)	)	-RRB-
cord-020871-1v6dcmt3	95	20	regularisation	regularisation	NN
cord-020871-1v6dcmt3	95	21	of	of	IN
cord-020871-1v6dcmt3	95	22	the	the	DT
cord-020871-1v6dcmt3	95	23	logistic	logistic	JJ
cord-020871-1v6dcmt3	95	24	regression	regression	NN
cord-020871-1v6dcmt3	95	25	classifier	classifier	NN
cord-020871-1v6dcmt3	95	26	on	on	IN
cord-020871-1v6dcmt3	95	27	the	the	DT
cord-020871-1v6dcmt3	95	28	accuracy	accuracy	NN
cord-020871-1v6dcmt3	95	29	,	,	,
cord-020871-1v6dcmt3	95	30	while	while	IN
cord-020871-1v6dcmt3	95	31	the	the	DT
cord-020871-1v6dcmt3	95	32	right	right	JJ
cord-020871-1v6dcmt3	95	33	panels	panel	NNS
cord-020871-1v6dcmt3	95	34	[	[	-LRB-
cord-020871-1v6dcmt3	95	35	(	(	-LRB-
cord-020871-1v6dcmt3	95	36	c	c	NN
cord-020871-1v6dcmt3	95	37	)	)	-RRB-
cord-020871-1v6dcmt3	95	38	and	and	CC
cord-020871-1v6dcmt3	95	39	(	(	-LRB-
cord-020871-1v6dcmt3	95	40	d	d	NN
cord-020871-1v6dcmt3	95	41	)	)	-RRB-
cord-020871-1v6dcmt3	95	42	]	]	-RRB-
cord-020871-1v6dcmt3	95	43	display	display	VB
cord-020871-1v6dcmt3	95	44	the	the	DT
cord-020871-1v6dcmt3	95	45	effect	effect	NN
cord-020871-1v6dcmt3	95	46	of	of	IN
cord-020871-1v6dcmt3	95	47	training	training	NN
cord-020871-1v6dcmt3	95	48	for	for	IN
cord-020871-1v6dcmt3	95	49	the	the	DT
cord-020871-1v6dcmt3	95	50	learning	learn	VBG
cord-020871-1v6dcmt3	95	51	algorithms	algorithm	NNS
cord-020871-1v6dcmt3	95	52	.	.	.
cord-020871-1v6dcmt3	96	1	The	the	DT
cord-020871-1v6dcmt3	96	2	two	two	CD
cord-020871-1v6dcmt3	96	3	symbols	symbol	NNS
cord-020871-1v6dcmt3	96	4	on	on	IN
cord-020871-1v6dcmt3	96	5	the	the	DT
cord-020871-1v6dcmt3	96	6	right	right	JJ
cord-020871-1v6dcmt3	96	7	axis	axis	NN
cord-020871-1v6dcmt3	96	8	in	in	IN
cord-020871-1v6dcmt3	96	9	panels	panel	NNS
cord-020871-1v6dcmt3	96	10	(	(	-LRB-
cord-020871-1v6dcmt3	96	11	a	a	LS
cord-020871-1v6dcmt3	96	12	)	)	-RRB-
cord-020871-1v6dcmt3	96	13	and	and	CC
cord-020871-1v6dcmt3	96	14	(	(	-LRB-
cord-020871-1v6dcmt3	96	15	b	b	LS
cord-020871-1v6dcmt3	96	16	)	)	-RRB-
cord-020871-1v6dcmt3	96	17	indicate	indicate	VBP
cord-020871-1v6dcmt3	96	18	the	the	DT
cord-020871-1v6dcmt3	96	19	best	good	JJS
cord-020871-1v6dcmt3	96	20	(	(	-LRB-
cord-020871-1v6dcmt3	96	21	FV	fv	NN
cord-020871-1v6dcmt3	96	22	-	-	HYPH
cord-020871-1v6dcmt3	96	23	moVMF	movmf	NN
cord-020871-1v6dcmt3	96	24	)	)	-RRB-
cord-020871-1v6dcmt3	96	25	results	result	NNS
cord-020871-1v6dcmt3	96	26	reported	report	VBN
cord-020871-1v6dcmt3	96	27	in	in	IN
cord-020871-1v6dcmt3	96	28	[	[	-LRB-
cord-020871-1v6dcmt3	96	29	14	14	CD
cord-020871-1v6dcmt3	96	30	]	]	-RRB-
cord-020871-1v6dcmt3	96	31	.	.	.
cord-020871-1v6dcmt3	96	32	 	 	_SP
cord-020871-1v6dcmt3	97	1	For	for	IN
cord-020871-1v6dcmt3	97	2	these	these	DT
cord-020871-1v6dcmt3	97	3	experiments	experiment	NNS
cord-020871-1v6dcmt3	97	4	,	,	,
cord-020871-1v6dcmt3	97	5	we	-PRON-	PRP
cord-020871-1v6dcmt3	97	6	extracted	extract	VBD
cord-020871-1v6dcmt3	97	7	from	from	IN
cord-020871-1v6dcmt3	97	8	every	every	DT
cord-020871-1v6dcmt3	97	9	document	document	NN
cord-020871-1v6dcmt3	97	10	of	of	IN
cord-020871-1v6dcmt3	97	11	the	the	DT
cord-020871-1v6dcmt3	97	12	test	test	NN
cord-020871-1v6dcmt3	97	13	collection	collection	NN
cord-020871-1v6dcmt3	97	14	all	all	PDT
cord-020871-1v6dcmt3	97	15	the	the	DT
cord-020871-1v6dcmt3	97	16	raw	raw	JJ
cord-020871-1v6dcmt3	97	17	text	text	NN
cord-020871-1v6dcmt3	97	18	,	,	,
cord-020871-1v6dcmt3	97	19	and	and	CC
cord-020871-1v6dcmt3	97	20	preprocessed	preprocesse	VBD
cord-020871-1v6dcmt3	97	21	it	-PRON-	PRP
cord-020871-1v6dcmt3	97	22	as	as	IN
cord-020871-1v6dcmt3	97	23	described	describe	VBN
cord-020871-1v6dcmt3	97	24	in	in	IN
cord-020871-1v6dcmt3	97	25	the	the	DT
cord-020871-1v6dcmt3	97	26	beginning	beginning	NN
cord-020871-1v6dcmt3	97	27	of	of	IN
cord-020871-1v6dcmt3	97	28	this	this	DT
cord-020871-1v6dcmt3	97	29	section	section	NN
cord-020871-1v6dcmt3	97	30	.	.	.
cord-020871-1v6dcmt3	98	1	The	the	DT
cord-020871-1v6dcmt3	98	2	documents	document	NNS
cord-020871-1v6dcmt3	98	3	were	be	VBD
cord-020871-1v6dcmt3	98	4	indexed	index	VBN
cord-020871-1v6dcmt3	98	5	and	and	CC
cord-020871-1v6dcmt3	98	6	retrieved	retrieve	VBN
cord-020871-1v6dcmt3	98	7	for	for	IN
cord-020871-1v6dcmt3	98	8	BM25	BM25	NNP
cord-020871-1v6dcmt3	98	9	with	with	IN
cord-020871-1v6dcmt3	98	10	the	the	DT
cord-020871-1v6dcmt3	98	11	Lucene	Lucene	NNP
cord-020871-1v6dcmt3	98	12	8.2	8.2	CD
cord-020871-1v6dcmt3	98	13	search	search	NN
cord-020871-1v6dcmt3	98	14	engine	engine	NN
cord-020871-1v6dcmt3	98	15	.	.	.
cord-020871-1v6dcmt3	99	1	We	-PRON-	PRP
cord-020871-1v6dcmt3	99	2	experimented	experiment	VBD
cord-020871-1v6dcmt3	99	3	with	with	IN
cord-020871-1v6dcmt3	99	4	three	three	CD
cord-020871-1v6dcmt3	99	5	topic	topic	NN
cord-020871-1v6dcmt3	99	6	processing	processing	NN
cord-020871-1v6dcmt3	99	7	ways	way	NNS
cord-020871-1v6dcmt3	99	8	:	:	:
cord-020871-1v6dcmt3	99	9	(	(	-LRB-
cord-020871-1v6dcmt3	99	10	1	1	LS
cord-020871-1v6dcmt3	99	11	)	)	-RRB-
cord-020871-1v6dcmt3	99	12	title	title	NN
cord-020871-1v6dcmt3	99	13	only	only	RB
cord-020871-1v6dcmt3	99	14	,	,	,
cord-020871-1v6dcmt3	99	15	(	(	-LRB-
cord-020871-1v6dcmt3	99	16	2	2	LS
cord-020871-1v6dcmt3	99	17	)	)	-RRB-
cord-020871-1v6dcmt3	99	18	description	description	NN
cord-020871-1v6dcmt3	99	19	only	only	RB
cord-020871-1v6dcmt3	99	20	,	,	,
cord-020871-1v6dcmt3	99	21	and	and	CC
cord-020871-1v6dcmt3	99	22	(	(	-LRB-
cord-020871-1v6dcmt3	99	23	3	3	LS
cord-020871-1v6dcmt3	99	24	)	)	-RRB-
cord-020871-1v6dcmt3	99	25	title	title	NN
cord-020871-1v6dcmt3	99	26	and	and	CC
cord-020871-1v6dcmt3	99	27	description	description	NN
cord-020871-1v6dcmt3	99	28	.	.	.
cord-020871-1v6dcmt3	100	1	The	the	DT
cord-020871-1v6dcmt3	100	2	third	third	JJ
cord-020871-1v6dcmt3	100	3	way	way	NN
cord-020871-1v6dcmt3	100	4	produces	produce	VBZ
cord-020871-1v6dcmt3	100	5	the	the	DT
cord-020871-1v6dcmt3	100	6	best	good	JJS
cord-020871-1v6dcmt3	100	7	results	result	NNS
cord-020871-1v6dcmt3	100	8	and	and	CC
cord-020871-1v6dcmt3	100	9	closest	close	JJS
cord-020871-1v6dcmt3	100	10	to	to	IN
cord-020871-1v6dcmt3	100	11	the	the	DT
cord-020871-1v6dcmt3	100	12	ones	one	NNS
cord-020871-1v6dcmt3	100	13	reported	report	VBN
cord-020871-1v6dcmt3	100	14	by	by	IN
cord-020871-1v6dcmt3	100	15	Zhang	Zhang	NNP
cord-020871-1v6dcmt3	100	16	et	et	NNP
cord-020871-1v6dcmt3	100	17	al	al	NNP
cord-020871-1v6dcmt3	100	18	.	.	.
cord-020871-1v6dcmt3	101	1	[	[	-LRB-
cord-020871-1v6dcmt3	101	2	14	14	CD
cord-020871-1v6dcmt3	101	3	]	]	-RRB-
cord-020871-1v6dcmt3	101	4	,	,	,
cord-020871-1v6dcmt3	101	5	and	and	CC
cord-020871-1v6dcmt3	101	6	hence	hence	RB
cord-020871-1v6dcmt3	101	7	are	be	VBP
cord-020871-1v6dcmt3	101	8	the	the	DT
cord-020871-1v6dcmt3	101	9	only	only	JJ
cord-020871-1v6dcmt3	101	10	ones	one	NNS
cord-020871-1v6dcmt3	101	11	reported	report	VBN
cord-020871-1v6dcmt3	101	12	here	here	RB
cord-020871-1v6dcmt3	101	13	.	.	.
cord-020871-1v6dcmt3	102	1	An	an	DT
cord-020871-1v6dcmt3	102	2	important	important	JJ
cord-020871-1v6dcmt3	102	3	aspect	aspect	NN
cord-020871-1v6dcmt3	102	4	of	of	IN
cord-020871-1v6dcmt3	102	5	BM25	BM25	NNP
cord-020871-1v6dcmt3	102	6	is	be	VBZ
cord-020871-1v6dcmt3	102	7	the	the	DT
cord-020871-1v6dcmt3	102	8	fact	fact	NN
cord-020871-1v6dcmt3	102	9	that	that	IN
cord-020871-1v6dcmt3	102	10	the	the	DT
cord-020871-1v6dcmt3	102	11	variation	variation	NN
cord-020871-1v6dcmt3	102	12	of	of	IN
cord-020871-1v6dcmt3	102	13	its	-PRON-	PRP$
cord-020871-1v6dcmt3	102	14	parameters	parameter	NNS
cord-020871-1v6dcmt3	102	15	k	k	NNP
cord-020871-1v6dcmt3	102	16	1	1	CD
cord-020871-1v6dcmt3	102	17	and	and	CC
cord-020871-1v6dcmt3	102	18	b	b	LS
cord-020871-1v6dcmt3	102	19	could	could	MD
cord-020871-1v6dcmt3	102	20	bring	bring	VB
cord-020871-1v6dcmt3	102	21	significant	significant	JJ
cord-020871-1v6dcmt3	102	22	improvement	improvement	NN
cord-020871-1v6dcmt3	102	23	in	in	IN
cord-020871-1v6dcmt3	102	24	performance	performance	NN
cord-020871-1v6dcmt3	102	25	,	,	,
cord-020871-1v6dcmt3	102	26	as	as	IN
cord-020871-1v6dcmt3	102	27	reported	report	VBN
cord-020871-1v6dcmt3	102	28	by	by	IN
cord-020871-1v6dcmt3	102	29	Lipani	Lipani	NNP
cord-020871-1v6dcmt3	102	30	et	et	NNP
cord-020871-1v6dcmt3	102	31	al	al	NNP
cord-020871-1v6dcmt3	102	32	.	.	.
cord-020871-1v6dcmt3	103	1	[	[	-LRB-
cord-020871-1v6dcmt3	103	2	8	8	CD
cord-020871-1v6dcmt3	103	3	]	]	-RRB-
cord-020871-1v6dcmt3	103	4	.	.	.
cord-020871-1v6dcmt3	104	1	Therefore	therefore	RB
cord-020871-1v6dcmt3	104	2	,	,	,
cord-020871-1v6dcmt3	104	3	we	-PRON-	PRP
cord-020871-1v6dcmt3	104	4	performed	perform	VBD
cord-020871-1v6dcmt3	104	5	a	a	DT
cord-020871-1v6dcmt3	104	6	parameter	parameter	NN
cord-020871-1v6dcmt3	104	7	scan	scan	NN
cord-020871-1v6dcmt3	104	8	for	for	IN
cord-020871-1v6dcmt3	104	9	k	k	NN
cord-020871-1v6dcmt3	104	10	1	1	CD
cord-020871-1v6dcmt3	104	11	∈	∈	NN
cord-020871-1v6dcmt3	104	12	[	[	-LRB-
cord-020871-1v6dcmt3	104	13	0	0	CD
cord-020871-1v6dcmt3	104	14	,	,	,
cord-020871-1v6dcmt3	104	15	3	3	CD
cord-020871-1v6dcmt3	104	16	]	]	-RRB-
cord-020871-1v6dcmt3	104	17	and	and	CC
cord-020871-1v6dcmt3	104	18	b	b	LS
cord-020871-1v6dcmt3	104	19	∈	∈	VB
cord-020871-1v6dcmt3	104	20	[	[	-LRB-
cord-020871-1v6dcmt3	104	21	0	0	CD
cord-020871-1v6dcmt3	104	22	,	,	,
cord-020871-1v6dcmt3	104	23	1	1	CD
cord-020871-1v6dcmt3	104	24	]	]	-RRB-
cord-020871-1v6dcmt3	104	25	with	with	IN
cord-020871-1v6dcmt3	104	26	a	a	DT
cord-020871-1v6dcmt3	104	27	0.05	0.05	CD
cord-020871-1v6dcmt3	104	28	step	step	NN
cord-020871-1v6dcmt3	104	29	size	size	NN
cord-020871-1v6dcmt3	104	30	for	for	IN
cord-020871-1v6dcmt3	104	31	both	both	DT
cord-020871-1v6dcmt3	104	32	parameters	parameter	NNS
cord-020871-1v6dcmt3	104	33	.	.	.
cord-020871-1v6dcmt3	105	1	For	for	IN
cord-020871-1v6dcmt3	105	2	every	every	DT
cord-020871-1v6dcmt3	105	3	TREC	TREC	NNP
cord-020871-1v6dcmt3	105	4	topic	topic	NN
cord-020871-1v6dcmt3	105	5	,	,	,
cord-020871-1v6dcmt3	105	6	the	the	DT
cord-020871-1v6dcmt3	105	7	scores	score	NNS
cord-020871-1v6dcmt3	105	8	of	of	IN
cord-020871-1v6dcmt3	105	9	the	the	DT
cord-020871-1v6dcmt3	105	10	top	top	JJ
cord-020871-1v6dcmt3	105	11	1000	1000	CD
cord-020871-1v6dcmt3	105	12	documents	document	NNS
cord-020871-1v6dcmt3	105	13	retrieved	retrieve	VBN
cord-020871-1v6dcmt3	105	14	from	from	IN
cord-020871-1v6dcmt3	105	15	BM25	BM25	NNP
cord-020871-1v6dcmt3	105	16	were	be	VBD
cord-020871-1v6dcmt3	105	17	normalised	normalise	VBN
cord-020871-1v6dcmt3	105	18	to	to	IN
cord-020871-1v6dcmt3	105	19	[	[	-LRB-
cord-020871-1v6dcmt3	105	20	0,1	0,1	CD
cord-020871-1v6dcmt3	105	21	]	]	-RRB-
cord-020871-1v6dcmt3	105	22	with	with	IN
cord-020871-1v6dcmt3	105	23	the	the	DT
cord-020871-1v6dcmt3	105	24	min	min	NNP
cord-020871-1v6dcmt3	105	25	-	-	HYPH
cord-020871-1v6dcmt3	105	26	max	max	NN
cord-020871-1v6dcmt3	105	27	normalisation	normalisation	NN
cord-020871-1v6dcmt3	105	28	method	method	NN
cord-020871-1v6dcmt3	105	29	,	,	,
cord-020871-1v6dcmt3	105	30	and	and	CC
cord-020871-1v6dcmt3	105	31	were	be	VBD
cord-020871-1v6dcmt3	105	32	used	use	VBN
cord-020871-1v6dcmt3	105	33	in	in	IN
cord-020871-1v6dcmt3	105	34	calculating	calculate	VBG
cord-020871-1v6dcmt3	105	35	the	the	DT
cord-020871-1v6dcmt3	105	36	scores	score	NNS
cord-020871-1v6dcmt3	105	37	of	of	IN
cord-020871-1v6dcmt3	105	38	the	the	DT
cord-020871-1v6dcmt3	105	39	documents	document	NNS
cord-020871-1v6dcmt3	105	40	for	for	IN
cord-020871-1v6dcmt3	105	41	the	the	DT
cord-020871-1v6dcmt3	105	42	combined	combined	JJ
cord-020871-1v6dcmt3	105	43	models	model	NNS
cord-020871-1v6dcmt3	105	44	[	[	-LRB-
cord-020871-1v6dcmt3	105	45	14	14	CD
cord-020871-1v6dcmt3	105	46	]	]	-RRB-
cord-020871-1v6dcmt3	105	47	.	.	.
cord-020871-1v6dcmt3	106	1	The	the	DT
cord-020871-1v6dcmt3	106	2	original	original	JJ
cord-020871-1v6dcmt3	106	3	results	result	NNS
cord-020871-1v6dcmt3	106	4	,	,	,
cord-020871-1v6dcmt3	106	5	those	those	DT
cord-020871-1v6dcmt3	106	6	of	of	IN
cord-020871-1v6dcmt3	106	7	our	-PRON-	PRP$
cord-020871-1v6dcmt3	106	8	replication	replication	NN
cord-020871-1v6dcmt3	106	9	experiments	experiment	NNS
cord-020871-1v6dcmt3	106	10	with	with	IN
cord-020871-1v6dcmt3	106	11	standard	standard	NN
cord-020871-1v6dcmt3	106	12	(	(	-LRB-
cord-020871-1v6dcmt3	106	13	k	k	NN
cord-020871-1v6dcmt3	106	14	1	1	CD
cord-020871-1v6dcmt3	106	15	=	=	SYM
cord-020871-1v6dcmt3	106	16	1.2	1.2	CD
cord-020871-1v6dcmt3	106	17	and	and	CC
cord-020871-1v6dcmt3	106	18	b	b	NN
cord-020871-1v6dcmt3	106	19	=	=	SYM
cord-020871-1v6dcmt3	106	20	0.75	0.75	CD
cord-020871-1v6dcmt3	106	21	)	)	-RRB-
cord-020871-1v6dcmt3	106	22	and	and	CC
cord-020871-1v6dcmt3	106	23	best	good	JJS
cord-020871-1v6dcmt3	106	24	BM25	BM25	NNP
cord-020871-1v6dcmt3	106	25	parameter	parameter	NN
cord-020871-1v6dcmt3	106	26	values	value	NNS
cord-020871-1v6dcmt3	106	27	-	-	HYPH
cord-020871-1v6dcmt3	106	28	measured	measure	VBN
cord-020871-1v6dcmt3	106	29	in	in	IN
cord-020871-1v6dcmt3	106	30	terms	term	NNS
cord-020871-1v6dcmt3	106	31	of	of	IN
cord-020871-1v6dcmt3	106	32	Mean	mean	JJ
cord-020871-1v6dcmt3	106	33	Average	Average	NNP
cord-020871-1v6dcmt3	106	34	Precision	Precision	NNP
cord-020871-1v6dcmt3	106	35	(	(	-LRB-
cord-020871-1v6dcmt3	106	36	MAP	MAP	NNP
cord-020871-1v6dcmt3	106	37	)	)	-RRB-
cord-020871-1v6dcmt3	106	38	and	and	CC
cord-020871-1v6dcmt3	106	39	Precision	Precision	NNP
cord-020871-1v6dcmt3	106	40	at	at	IN
cord-020871-1v6dcmt3	106	41	20	20	CD
cord-020871-1v6dcmt3	107	1	(	(	-LRB-
cord-020871-1v6dcmt3	107	2	P@20)-are	p@20)-are	NN
cord-020871-1v6dcmt3	107	3	outlined	outline	VBN
cord-020871-1v6dcmt3	107	4	in	in	IN
cord-020871-1v6dcmt3	107	5	Table	Table	NNP
cord-020871-1v6dcmt3	107	6	3	3	CD
cord-020871-1v6dcmt3	107	7	.	.	.
cord-020871-1v6dcmt3	108	1	We	-PRON-	PRP
cord-020871-1v6dcmt3	108	2	replicated	replicate	VBD
cord-020871-1v6dcmt3	108	3	previously	previously	RB
cord-020871-1v6dcmt3	108	4	reported	report	VBD
cord-020871-1v6dcmt3	108	5	experiments	experiment	NNS
cord-020871-1v6dcmt3	108	6	that	that	WDT
cord-020871-1v6dcmt3	108	7	presented	present	VBD
cord-020871-1v6dcmt3	108	8	evidence	evidence	NN
cord-020871-1v6dcmt3	108	9	that	that	IN
cord-020871-1v6dcmt3	108	10	a	a	DT
cord-020871-1v6dcmt3	108	11	new	new	JJ
cord-020871-1v6dcmt3	108	12	mixture	mixture	NN
cord-020871-1v6dcmt3	108	13	model	model	NN
cord-020871-1v6dcmt3	108	14	,	,	,
cord-020871-1v6dcmt3	108	15	based	base	VBN
cord-020871-1v6dcmt3	108	16	on	on	IN
cord-020871-1v6dcmt3	108	17	von	von	NNP
cord-020871-1v6dcmt3	108	18	Mises	Mises	NNP
cord-020871-1v6dcmt3	108	19	-	-	HYPH
cord-020871-1v6dcmt3	108	20	Fisher	Fisher	NNP
cord-020871-1v6dcmt3	108	21	distributions	distribution	NNS
cord-020871-1v6dcmt3	108	22	,	,	,
cord-020871-1v6dcmt3	108	23	outperformed	outperform	VBD
cord-020871-1v6dcmt3	108	24	a	a	DT
cord-020871-1v6dcmt3	108	25	series	series	NN
cord-020871-1v6dcmt3	108	26	of	of	IN
cord-020871-1v6dcmt3	108	27	other	other	JJ
cord-020871-1v6dcmt3	108	28	models	model	NNS
cord-020871-1v6dcmt3	108	29	in	in	IN
cord-020871-1v6dcmt3	108	30	three	three	CD
cord-020871-1v6dcmt3	108	31	tasks	task	NNS
cord-020871-1v6dcmt3	108	32	(	(	-LRB-
cord-020871-1v6dcmt3	108	33	classification	classification	NN
cord-020871-1v6dcmt3	108	34	,	,	,
cord-020871-1v6dcmt3	108	35	clustering	clustering	NN
cord-020871-1v6dcmt3	108	36	,	,	,
cord-020871-1v6dcmt3	108	37	and	and	CC
cord-020871-1v6dcmt3	108	38	retrievalwhen	retrievalwhen	NNP
cord-020871-1v6dcmt3	108	39	combined	combine	VBN
cord-020871-1v6dcmt3	108	40	with	with	IN
cord-020871-1v6dcmt3	108	41	standard	standard	JJ
cord-020871-1v6dcmt3	108	42	retrieval	retrieval	NN
cord-020871-1v6dcmt3	108	43	models	model	NNS
cord-020871-1v6dcmt3	108	44	)	)	-RRB-
cord-020871-1v6dcmt3	108	45	.	.	.
cord-020871-1v6dcmt3	109	1	Since	since	IN
cord-020871-1v6dcmt3	109	2	the	the	DT
cord-020871-1v6dcmt3	109	3	source	source	NN
cord-020871-1v6dcmt3	109	4	code	code	NN
cord-020871-1v6dcmt3	109	5	was	be	VBD
cord-020871-1v6dcmt3	109	6	not	not	RB
cord-020871-1v6dcmt3	109	7	released	release	VBN
cord-020871-1v6dcmt3	109	8	in	in	IN
cord-020871-1v6dcmt3	109	9	the	the	DT
cord-020871-1v6dcmt3	109	10	original	original	JJ
cord-020871-1v6dcmt3	109	11	paper	paper	NN
cord-020871-1v6dcmt3	109	12	,	,	,
cord-020871-1v6dcmt3	109	13	important	important	JJ
cord-020871-1v6dcmt3	109	14	implementation	implementation	NN
cord-020871-1v6dcmt3	109	15	and	and	CC
cord-020871-1v6dcmt3	109	16	formulation	formulation	NN
cord-020871-1v6dcmt3	109	17	details	detail	NNS
cord-020871-1v6dcmt3	109	18	were	be	VBD
cord-020871-1v6dcmt3	109	19	omitted	omit	VBN
cord-020871-1v6dcmt3	109	20	,	,	,
cord-020871-1v6dcmt3	109	21	and	and	CC
cord-020871-1v6dcmt3	109	22	the	the	DT
cord-020871-1v6dcmt3	109	23	authors	author	NNS
cord-020871-1v6dcmt3	109	24	never	never	RB
cord-020871-1v6dcmt3	109	25	replied	reply	VBD
cord-020871-1v6dcmt3	109	26	to	to	IN
cord-020871-1v6dcmt3	109	27	our	-PRON-	PRP$
cord-020871-1v6dcmt3	109	28	request	request	NN
cord-020871-1v6dcmt3	109	29	for	for	IN
cord-020871-1v6dcmt3	109	30	information	information	NN
cord-020871-1v6dcmt3	109	31	,	,	,
cord-020871-1v6dcmt3	109	32	a	a	DT
cord-020871-1v6dcmt3	109	33	significant	significant	JJ
cord-020871-1v6dcmt3	109	34	effort	effort	NN
cord-020871-1v6dcmt3	109	35	has	have	VBZ
cord-020871-1v6dcmt3	109	36	been	be	VBN
cord-020871-1v6dcmt3	109	37	devoted	devote	VBN
cord-020871-1v6dcmt3	109	38	to	to	TO
cord-020871-1v6dcmt3	109	39	reverse	reverse	VB
cord-020871-1v6dcmt3	109	40	engineer	engineer	NN
cord-020871-1v6dcmt3	109	41	the	the	DT
cord-020871-1v6dcmt3	109	42	experiments	experiment	NNS
cord-020871-1v6dcmt3	109	43	.	.	.
cord-020871-1v6dcmt3	110	1	In	in	IN
cord-020871-1v6dcmt3	110	2	general	general	JJ
cord-020871-1v6dcmt3	110	3	,	,	,
cord-020871-1v6dcmt3	110	4	for	for	IN
cord-020871-1v6dcmt3	110	5	none	none	NN
cord-020871-1v6dcmt3	110	6	of	of	IN
cord-020871-1v6dcmt3	110	7	the	the	DT
cord-020871-1v6dcmt3	110	8	tasks	task	NNS
cord-020871-1v6dcmt3	110	9	were	be	VBD
cord-020871-1v6dcmt3	110	10	we	-PRON-	PRP
cord-020871-1v6dcmt3	110	11	able	able	JJ
cord-020871-1v6dcmt3	110	12	to	to	TO
cord-020871-1v6dcmt3	110	13	confirm	confirm	VB
cord-020871-1v6dcmt3	110	14	the	the	DT
cord-020871-1v6dcmt3	110	15	conclusions	conclusion	NNS
cord-020871-1v6dcmt3	110	16	of	of	IN
cord-020871-1v6dcmt3	110	17	the	the	DT
cord-020871-1v6dcmt3	110	18	previous	previous	JJ
cord-020871-1v6dcmt3	110	19	experiments	experiment	NNS
cord-020871-1v6dcmt3	110	20	:	:	:
cord-020871-1v6dcmt3	110	21	we	-PRON-	PRP
cord-020871-1v6dcmt3	110	22	do	do	VBP
cord-020871-1v6dcmt3	110	23	not	not	RB
cord-020871-1v6dcmt3	110	24	have	have	VB
cord-020871-1v6dcmt3	110	25	enough	enough	JJ
cord-020871-1v6dcmt3	110	26	evidence	evidence	NN
cord-020871-1v6dcmt3	110	27	to	to	TO
cord-020871-1v6dcmt3	110	28	conclude	conclude	VB
cord-020871-1v6dcmt3	110	29	that	that	IN
cord-020871-1v6dcmt3	110	30	FV	fv	NN
cord-020871-1v6dcmt3	110	31	-	-	HYPH
cord-020871-1v6dcmt3	110	32	moVMF	movmf	NN
cord-020871-1v6dcmt3	110	33	outperforms	outperform	VBZ
cord-020871-1v6dcmt3	110	34	the	the	DT
cord-020871-1v6dcmt3	110	35	other	other	JJ
cord-020871-1v6dcmt3	110	36	methods	method	NNS
cord-020871-1v6dcmt3	110	37	.	.	.
cord-020871-1v6dcmt3	111	1	The	the	DT
cord-020871-1v6dcmt3	111	2	situation	situation	NN
cord-020871-1v6dcmt3	111	3	is	be	VBZ
cord-020871-1v6dcmt3	111	4	rather	rather	RB
cord-020871-1v6dcmt3	111	5	different	different	JJ
cord-020871-1v6dcmt3	111	6	when	when	WRB
cord-020871-1v6dcmt3	111	7	considering	consider	VBG
cord-020871-1v6dcmt3	111	8	the	the	DT
cord-020871-1v6dcmt3	111	9	effectiveness	effectiveness	NN
cord-020871-1v6dcmt3	111	10	of	of	IN
cord-020871-1v6dcmt3	111	11	these	these	DT
cord-020871-1v6dcmt3	111	12	document	document	NN
cord-020871-1v6dcmt3	111	13	representations	representation	NNS
cord-020871-1v6dcmt3	111	14	for	for	IN
cord-020871-1v6dcmt3	111	15	clustering	clustering	NN
cord-020871-1v6dcmt3	111	16	purposes	purpose	NNS
cord-020871-1v6dcmt3	111	17	:	:	:
cord-020871-1v6dcmt3	111	18	we	-PRON-	PRP
cord-020871-1v6dcmt3	111	19	find	find	VBP
cord-020871-1v6dcmt3	111	20	indeed	indeed	RB
cord-020871-1v6dcmt3	111	21	that	that	IN
cord-020871-1v6dcmt3	111	22	the	the	DT
cord-020871-1v6dcmt3	111	23	FV	FV	NNP
cord-020871-1v6dcmt3	111	24	-	-	HYPH
cord-020871-1v6dcmt3	111	25	moVMF	movmf	NN
cord-020871-1v6dcmt3	111	26	significantly	significantly	RB
cord-020871-1v6dcmt3	111	27	underperforms	underperform	VBZ
cord-020871-1v6dcmt3	111	28	,	,	,
cord-020871-1v6dcmt3	111	29	contradicting	contradict	VBG
cord-020871-1v6dcmt3	111	30	previous	previous	JJ
cord-020871-1v6dcmt3	111	31	conclusions	conclusion	NNS
cord-020871-1v6dcmt3	111	32	.	.	.
cord-020871-1v6dcmt3	112	1	In	in	IN
cord-020871-1v6dcmt3	112	2	the	the	DT
cord-020871-1v6dcmt3	112	3	case	case	NN
cord-020871-1v6dcmt3	112	4	of	of	IN
cord-020871-1v6dcmt3	112	5	retrieval	retrieval	NN
cord-020871-1v6dcmt3	112	6	,	,	,
cord-020871-1v6dcmt3	112	7	although	although	IN
cord-020871-1v6dcmt3	112	8	Zhang	Zhang	NNP
cord-020871-1v6dcmt3	112	9	et	et	NNP
cord-020871-1v6dcmt3	112	10	al	al	NNP
cord-020871-1v6dcmt3	112	11	.	.	NNP
cord-020871-1v6dcmt3	112	12	's	's	POS
cord-020871-1v6dcmt3	112	13	proposed	propose	VBN
cord-020871-1v6dcmt3	112	14	method	method	NN
cord-020871-1v6dcmt3	112	15	(	(	-LRB-
cord-020871-1v6dcmt3	112	16	FV	fv	NN
cord-020871-1v6dcmt3	112	17	-	-	HYPH
cord-020871-1v6dcmt3	112	18	moVMF	movmf	NN
cord-020871-1v6dcmt3	112	19	)	)	-RRB-
cord-020871-1v6dcmt3	112	20	indeed	indeed	RB
cord-020871-1v6dcmt3	112	21	boosts	boost	VBZ
cord-020871-1v6dcmt3	112	22	BM25	BM25	NNP
cord-020871-1v6dcmt3	112	23	,	,	,
cord-020871-1v6dcmt3	112	24	it	-PRON-	PRP
cord-020871-1v6dcmt3	112	25	does	do	VBZ
cord-020871-1v6dcmt3	112	26	not	not	RB
cord-020871-1v6dcmt3	112	27	outperform	outperform	VB
cord-020871-1v6dcmt3	112	28	most	most	JJS
cord-020871-1v6dcmt3	112	29	of	of	IN
cord-020871-1v6dcmt3	112	30	the	the	DT
cord-020871-1v6dcmt3	112	31	other	other	JJ
cord-020871-1v6dcmt3	112	32	models	model	NNS
cord-020871-1v6dcmt3	112	33	it	-PRON-	PRP
cord-020871-1v6dcmt3	112	34	was	be	VBD
cord-020871-1v6dcmt3	112	35	compared	compare	VBN
cord-020871-1v6dcmt3	112	36	to	to	IN
cord-020871-1v6dcmt3	112	37	.	.	.
cord-020871-1v6dcmt3	113	1	Clustering	cluster	VBG
cord-020871-1v6dcmt3	113	2	on	on	IN
cord-020871-1v6dcmt3	113	3	the	the	DT
cord-020871-1v6dcmt3	113	4	unit	unit	NN
cord-020871-1v6dcmt3	113	5	hypersphere	hypersphere	NN
cord-020871-1v6dcmt3	113	6	using	use	VBG
cord-020871-1v6dcmt3	113	7	von	von	NNP
cord-020871-1v6dcmt3	113	8	Mises	Mises	NNP
cord-020871-1v6dcmt3	113	9	-	-	HYPH
cord-020871-1v6dcmt3	113	10	Fisher	Fisher	NNP
cord-020871-1v6dcmt3	113	11	distributions	distribution	NNS
cord-020871-1v6dcmt3	113	12	Latent	Latent	NNP
cord-020871-1v6dcmt3	113	13	Dirichlet	dirichlet	JJ
cord-020871-1v6dcmt3	113	14	allocation	allocation	NN
cord-020871-1v6dcmt3	113	15	Aggregating	aggregate	VBG
cord-020871-1v6dcmt3	113	16	continuous	continuous	JJ
cord-020871-1v6dcmt3	113	17	word	word	NN
cord-020871-1v6dcmt3	113	18	embeddings	embedding	NNS
cord-020871-1v6dcmt3	113	19	for	for	IN
cord-020871-1v6dcmt3	113	20	information	information	NN
cord-020871-1v6dcmt3	113	21	retrieval	retrieval	NN
cord-020871-1v6dcmt3	113	22	Indexing	indexing	NN
cord-020871-1v6dcmt3	113	23	by	by	IN
cord-020871-1v6dcmt3	113	24	latent	latent	JJ
cord-020871-1v6dcmt3	113	25	semantic	semantic	JJ
cord-020871-1v6dcmt3	113	26	analysis	analysis	NN
cord-020871-1v6dcmt3	113	27	Distributional	distributional	JJ
cord-020871-1v6dcmt3	113	28	structure	structure	NN
cord-020871-1v6dcmt3	113	29	.	.	.
cord-020871-1v6dcmt3	114	1	Word	word	NN
cord-020871-1v6dcmt3	114	2	Let	let	VB
cord-020871-1v6dcmt3	115	1	's	's	POS
cord-020871-1v6dcmt3	115	2	measure	measure	NN
cord-020871-1v6dcmt3	115	3	run	run	VBN
cord-020871-1v6dcmt3	115	4	time	time	NN
cord-020871-1v6dcmt3	115	5	!	!	.
cord-020871-1v6dcmt3	116	1	Extending	extend	VBG
cord-020871-1v6dcmt3	116	2	the	the	DT
cord-020871-1v6dcmt3	116	3	IR	IR	NNP
cord-020871-1v6dcmt3	116	4	replicability	replicability	NN
cord-020871-1v6dcmt3	116	5	infrastructure	infrastructure	NN
cord-020871-1v6dcmt3	116	6	to	to	TO
cord-020871-1v6dcmt3	116	7	include	include	VB
cord-020871-1v6dcmt3	116	8	performance	performance	NN
cord-020871-1v6dcmt3	116	9	aspects	aspect	NNS
cord-020871-1v6dcmt3	117	1	Distributed	distribute	VBN
cord-020871-1v6dcmt3	117	2	representations	representation	NNS
cord-020871-1v6dcmt3	117	3	of	of	IN
cord-020871-1v6dcmt3	117	4	sentences	sentence	NNS
cord-020871-1v6dcmt3	117	5	and	and	CC
cord-020871-1v6dcmt3	117	6	documents	document	NNS
cord-020871-1v6dcmt3	117	7	Verboseness	verboseness	JJ
cord-020871-1v6dcmt3	117	8	fission	fission	NN
cord-020871-1v6dcmt3	117	9	for	for	IN
cord-020871-1v6dcmt3	118	1	BM25	BM25	NNP
cord-020871-1v6dcmt3	118	2	document	document	NN
cord-020871-1v6dcmt3	118	3	length	length	NN
cord-020871-1v6dcmt3	118	4	normalization	normalization	NN
cord-020871-1v6dcmt3	119	1	Efficient	efficient	JJ
cord-020871-1v6dcmt3	119	2	estimation	estimation	NN
cord-020871-1v6dcmt3	119	3	of	of	IN
cord-020871-1v6dcmt3	119	4	word	word	NN
cord-020871-1v6dcmt3	119	5	representations	representation	NNS
cord-020871-1v6dcmt3	119	6	in	in	IN
cord-020871-1v6dcmt3	119	7	vector	vector	NN
cord-020871-1v6dcmt3	119	8	space	space	NN
cord-020871-1v6dcmt3	119	9	Neural	Neural	NNP
cord-020871-1v6dcmt3	119	10	information	information	NN
cord-020871-1v6dcmt3	119	11	retrieval	retrieval	NN
cord-020871-1v6dcmt3	119	12	:	:	:
cord-020871-1v6dcmt3	119	13	at	at	IN
cord-020871-1v6dcmt3	119	14	the	the	DT
cord-020871-1v6dcmt3	119	15	end	end	NN
cord-020871-1v6dcmt3	119	16	of	of	IN
cord-020871-1v6dcmt3	119	17	the	the	DT
cord-020871-1v6dcmt3	119	18	early	early	JJ
cord-020871-1v6dcmt3	119	19	years	year	NNS
cord-020871-1v6dcmt3	119	20	A	a	DT
cord-020871-1v6dcmt3	119	21	sentimental	sentimental	JJ
cord-020871-1v6dcmt3	119	22	education	education	NN
cord-020871-1v6dcmt3	119	23	:	:	:
cord-020871-1v6dcmt3	119	24	sentiment	sentiment	NN
cord-020871-1v6dcmt3	119	25	analysis	analysis	NN
cord-020871-1v6dcmt3	119	26	using	use	VBG
cord-020871-1v6dcmt3	119	27	subjectivity	subjectivity	NN
cord-020871-1v6dcmt3	119	28	summarization	summarization	NN
cord-020871-1v6dcmt3	119	29	based	base	VBN
cord-020871-1v6dcmt3	119	30	on	on	IN
cord-020871-1v6dcmt3	119	31	minimum	minimum	JJ
cord-020871-1v6dcmt3	119	32	cuts	cut	NNS
cord-020871-1v6dcmt3	119	33	Seeing	see	VBG
cord-020871-1v6dcmt3	119	34	stars	star	NNS
cord-020871-1v6dcmt3	119	35	:	:	:
cord-020871-1v6dcmt3	119	36	exploiting	exploit	VBG
cord-020871-1v6dcmt3	119	37	class	class	NN
cord-020871-1v6dcmt3	119	38	relationships	relationship	NNS
cord-020871-1v6dcmt3	119	39	for	for	IN
cord-020871-1v6dcmt3	119	40	sentiment	sentiment	NN
cord-020871-1v6dcmt3	119	41	categorization	categorization	NN
cord-020871-1v6dcmt3	119	42	with	with	IN
cord-020871-1v6dcmt3	119	43	respect	respect	NN
cord-020871-1v6dcmt3	119	44	to	to	IN
cord-020871-1v6dcmt3	119	45	rating	rating	NN
cord-020871-1v6dcmt3	119	46	scales	scale	NNS
cord-020871-1v6dcmt3	120	1	The	the	DT
cord-020871-1v6dcmt3	120	2	TREC	TREC	NNP
cord-020871-1v6dcmt3	120	3	robust	robust	JJ
cord-020871-1v6dcmt3	120	4	retrieval	retrieval	NN
cord-020871-1v6dcmt3	120	5	track	track	NN
cord-020871-1v6dcmt3	120	6	.	.	.
cord-020871-1v6dcmt3	121	1	SIGIR	SIGIR	NNP
cord-020871-1v6dcmt3	121	2	Forum	Forum	NNP
cord-020871-1v6dcmt3	122	1	Aggregating	aggregate	VBG
cord-020871-1v6dcmt3	122	2	neural	neural	JJ
cord-020871-1v6dcmt3	122	3	word	word	NN
cord-020871-1v6dcmt3	122	4	embeddings	embedding	NNS
cord-020871-1v6dcmt3	122	5	for	for	IN
cord-020871-1v6dcmt3	122	6	document	document	NN
cord-020871-1v6dcmt3	122	7	representation	representation	NN
cord-020871-1v6dcmt3	122	8	EARS	ear	VBZ
cord-020871-1v6dcmt3	122	9	2019	2019	CD
cord-020871-1v6dcmt3	122	10	:	:	:
cord-020871-1v6dcmt3	123	1	the	the	DT
cord-020871-1v6dcmt3	123	2	2nd	2nd	JJ
cord-020871-1v6dcmt3	123	3	international	international	JJ
cord-020871-1v6dcmt3	123	4	workshop	workshop	NN
cord-020871-1v6dcmt3	123	5	on	on	IN
cord-020871-1v6dcmt3	123	6	explainable	explainable	JJ
cord-020871-1v6dcmt3	123	7	recommendation	recommendation	NN
cord-020871-1v6dcmt3	123	8	and	and	CC
cord-020871-1v6dcmt3	123	9	search	search	NN
cord-020871-1v6dcmt3	123	10	Authors	author	NNS
cord-020871-1v6dcmt3	123	11	are	be	VBP
cord-020871-1v6dcmt3	123	12	partially	partially	RB
cord-020871-1v6dcmt3	123	13	supported	support	VBN
cord-020871-1v6dcmt3	123	14	by	by	IN
cord-020871-1v6dcmt3	123	15	the	the	DT
cord-020871-1v6dcmt3	123	16	H2020	H2020	NNP
cord-020871-1v6dcmt3	123	17	Safe	Safe	NNP
cord-020871-1v6dcmt3	123	18	-	-	HYPH
cord-020871-1v6dcmt3	123	19	DEED	deed	NN
cord-020871-1v6dcmt3	123	20	project	project	NN
cord-020871-1v6dcmt3	123	21	(	(	-LRB-
cord-020871-1v6dcmt3	123	22	GA	GA	NNP
cord-020871-1v6dcmt3	123	23	825225	825225	CD
cord-020871-1v6dcmt3	123	24	)	)	-RRB-
cord-020871-1v6dcmt3	123	25	.	.	.