Distant Reader - journal-advancesInInformationRetrieval-cord

Introduction

This is a Distant Reader "study carrel", a set of structured data intended to help the student, researcher, or scholar use & understand a corpus.

This study carrel was created on 2021-05-30 by Eric Morgan <emorgan@nd.edu>. The carrel was created using the Distant Reader cord process, and the input was the result of a query applied to a local mirror of CORD, a data set of scholarly articles on the topic of COVID-19. The actual query was: facet_journal:"Advances in Information Retrieval". The results of this query were saved in a cache and transformed into a set of plain text files. All of the analysis -- "reading" -- has been done against these plain text files. For example, a short narrative report has been created. This Web page is a more verbose version of that report.

All study carrels are self-contained -- no Internet connection is necessary to use them. Download this carrel for offline reading. The carrel is made up of many subdirectories and data files. The manifest describes each one in greater detail.

Size

There are 43 item(s) in this carrel, and this carrel is 100,167 words long. Each item in your study carrel is, on average, 3,338 words long. If you dig deeper, then you might want to save yourself some time by reading a shorter item. On the other hand, if your desire is for more detail, then you might consider reading a longer item. The following charts illustrate the overall size of the carrel.

Readability

On a scale from 0 to 100, where 0 is very difficult and 100 is very easy, the documents have an average readability score of 55. Consequently, if you want to read something more simplistic, then consider a document with a higher score. If you want something more specialized, then consider something with a lower score. The following charts illustrate the overall readability of the carrel.

Word Frequencies

By merely counting & tabulating the frequency of individual words or phrases, you can begin to get an understanding of the carrel's "aboutness". Excluding "stop words", some of the more frequent words include:

model, query, information, text, based, i, retrieval, document, data, task, models, using, used, learning, results, set, use, dataset, also, documents, word, different, two, one, search, words, embedding, user, neural, image, tasks, attention, evaluation, approach, features, performance, work, training, analysis, ranking, number, queries, graph, users, network, proposed, methods, embeddings, relevance, term

Using the three most frequent words, the three files containing all of those words the most are Relevance Ranking Based on Query-Aware Context Analysis, What Can Task Teach Us About Query Reformulations?, and Learning to Rank Images with Cross-Modal Graph Convolutions.

The most frequent two-word phrases (bigrams) include:

information retrieval, seed words, social media, word embeddings, neural networks, query terms, schema labels, question answering, sentiment analysis, query term, cord uid, retrieval doi, neural network, doc id, schema label, recommender systems, word embedding, training data, data sets, retrieval models, clef ehealth, bantu languages, attention layer, deep learning, contact recommendation, natural language, proposed model, attention mechanism, test set, training set, embedding space, keyphrase extraction, data set, knowledge graph, posting lists, topic modeling, convolutional neural, local context, language models, review embedding, sentiment lexicon, search engines, language model, modal retrieval, future work, query expansion, matrix factorization, web search, semantic matching, previous work

And the three file that use all of the three most frequent phrases are Seed-Guided Deep Document Clustering Relevance Ranking Based on Query-Aware Context Analysis, and Axiomatic Analysis of Contact Recommendation Methods in Social Networks: An IR Perspective.

While often deemed superficial or sophomoric, rudimentary frequencies and their associated "word clouds" can be quite insightful:

Keywords

Sets of keywords -- statistically significant words -- can be enumerated by comparing the relative frequency of words with the number of times the words appear in an entire corpus. Some of the most statistically significant keywords in the carrel include:

user, image, task, query, document, word, review, model, lucene, graph, english, dataset, claim, bm25, vrss, view, tweet, trec, topic, text, term, system, symptom, suel, session, sentence, seed, sd2c, schema, runyankore, recommendation, ranker, question, product, prf, premise, patent, passage, ontology, node, network, location, lmp, lmote, list, lexicon, language, label, item, irony

And now word clouds really begin to shine:

Topic Modeling

Topic modeling is another popular approach to connoting the aboutness of a corpus. If the study carrel could be summed up in a single word, then that word might be model, and CLEF eHealth Evaluation Lab 2020 is most about that word.

If the study carrel could be summed up in three words ("topics") then those words and their significantly associated titles include:

If the study carrel could be summed up in five topics, and each topic were each denoted with three words, then those topics and their most significantly associated files would be:

model, document, text - Learning to Rank Images with Cross-Modal Graph Convolutions
query, task, retrieval - What Can Task Teach Us About Query Reformulations?
data, learning, tweets - Counterfactual Online Learning to Rank
schema, task, dataset - bias goggles: Graph-Based Computation of the Bias of Web Domains Through the Eyes of Users
graph, nodes, information - KvGR: A Graph-Based Interface for Explorative Sequential Question Answering on Heterogeneous Information Sources

Moreover, the totality of the study carrel's aboutness, can be visualized with the following pie chart:

Noun & Verbs

Through an analysis of your study carrel's parts-of-speech, you are able to answer question beyonds aboutness. For example, a list of the most frequent nouns helps you answer what questions; "What is discussed in this collection?":

model, query, information, text, document, data, models, retrieval, task, results, documents, user, dataset, word, words, image, search, users, attention, approach, tasks, performance, work, number, set, queries, features, learning, methods, graph, embeddings, terms, training, representations, evaluation, system, review, term, analysis, network, context, language, images, relevance, approaches, representation, similarity, sentiment, networks, datasets

An enumeration of the verbs helps you learn what actions take place in a text or what the things in the text do. Very frequently, the most common lemmatized verbs are "be", "have", and "do"; the more interesting verbs usually occur further down the list of frequencies:

using, based, learns, show, propose, consider, provided, given, follows, generate, made, training, include, ranking, evaluate, compared, setting, finding, compute, contain, embedding, define, described, performs, obtain, supporting, retrieve, represent, identify, saw, introduce, predicts, improving, related, present, existing, takes, applying, extract, selected, combined, focused, capture, reported, require, outperforms, needs, sharing, observed, denotes

Proper Nouns

An extraction of proper nouns helps you determine the names of people and places in your study carrel.

IR, Sect, BM25, Table, Fig, Eq, Retrieval, S, Lucene, Information, English, K, Twitter, COLTR, D, T, Bantu, BERT, i, DOI, TREC, Neural, TransRev, sha, Task, M, L, eRisk, C, VRSS, F, BC, LSTM, CNN, A, Wikipedia, Analysis, Network, LDA, dom, W, AUC, TF, Model, Cond, Amazon, corpus, IDF, DMP, Adam

An analysis of personal pronouns enables you to answer at least two questions: 1) "What, if any, is the overall gender of my study carrel?", and 2) "To what degree are the texts in my study carrel self-centered versus inclusive?"

we, our, it, their, they, i, its, them, one, us, you, he, itself, my, his, your, u, ours, me, she, s, ourselves, themselves, ndcg@10, mine, him, her, 's, Π, f

Below are words cloud of your study carrel's proper & personal pronouns.

Adjectives & Verbs

Learning about a corpus's adjectives and adverbs helps you answer how questions: "How are things described and how are things done?" An analysis of adjectives and adverbs also points to a corpus's overall sentiment. "In general, is my study carrel positive or negative?"

different, neural, semantic, new, similar, relevant, large, deep, previous, first, social, specific, available, non, multi, many, single, best, modal, cross, original, several, common, online, multiple, important, local, better, standard, simple, second, long, high, final, effective, top, various, visual, additional, real, able, latent, traditional, automatic, small, natural, possible, biased, average, textual

also, however, therefore, well, first, respectively, finally, even, instead, significantly, recently, often, better, furthermore, directly, still, automatically, fully, specifically, rather, especially, additionally, hence, randomly, much, moreover, always, already, usually, less, generally, widely, together, similarly, publicly, previously, now, typically, particularly, manually, semantically, far, otherwise, mainly, actually, simply, namely, jointly, highly, effectively

Next steps

There is much more to a study carrel than the things outlined above. Use this page's menubar to navigate and explore in more detail. There you will find additional features & functions including: ngrams, parts-of-speech, grammars, named entities, topic modeling, a simple search interface, etc.

Again, study carrels are self-contained. Download this carrel for offline viewing and use.

Thank you for using the Distant Reader.