id	author	title	date	pages	extension	mime	words	sentences	flesch	summary	cache	txt
work_m5ynmpp4jnfqfaura642udcc4a	Victoria Zayats	Conversation Modeling on Reddit Using a Graph-Structured LSTM	2018	12	.pdf	application/pdf	6989	558	65	Figure 2: An example of model propagation in a graph-structured LSTM. endorsement (Fang et al., 2016), is the task of interest in our work on tree-structured modeling of discussions. structure and timing are important in predicting popularity (Fang et al., 2016), the LSTM units include et al., 2016), our model makes use of the full discussion thread in predicting popularity. By introducing a forward-backward treestructured model, we provide a mechanism for leveraging early responses in predicting popularity, as with LSTMs; evaluation of the model on the popularity prediction task using Reddit discussions; and that characterizes a full threaded discussion, assuming a tree-structured response network and accounting for the relative order of the comments. The comment text features, denoted xct, are generated using a simple average bag-of-words representation learned during the training: bidirectional graph-LSTM model, language is helping identify overpredicted cases more than underpredicted ones.	./cache/work_m5ynmpp4jnfqfaura642udcc4a.pdf	./txt/work_m5ynmpp4jnfqfaura642udcc4a.txt