Modeling Concept Dependencies in a Scientific Corpus

Abstract

Our goal is to generate reading lists for students that help them optimally learn technical material. Existing retrieval algorithms return items directly relevant to a query but do not return results to help users read about the concepts supporting their query. This is because the dependency structure of concepts that must be understood before reading material pertaining to a given query is never considered. Here we formulate an information-theoretic view of concept dependency and present methods to construct a "concept graph" automatically from a text corpus. We perform the first human evaluation of concept dependency edges (to be published as open data), and the results verify the feasibility of automatic approaches for inferring concepts and their dependency relations. This result can support search capabilities that may be tuned to help users learn a subject rather than retrieve documents based on a single query.

TutorialBank: A Manually-Collected Corpus for Prerequisite Chains, Survey Extraction and Resource Recommendation

“…They introduce both a synthetic dataset as well as one scraped from 11 universities which includes course prerequisites as well as conceptprerequisite labels. Concept graphs are also used in (Gordon et al, 2016) to address the problem of developing reading lists for students. The concept graph in this case is a labeled graph where nodes represent both documents and concepts (determined using Latent Dirichlet Allocation (LDA) (Blei et al, 2003)), and edges represent dependencies.…”

“…We use similar categories for classifying pedagogical function as , but our corpus is hand-picked and over four-times larger, while exhibiting similar annotation agreement. Gordon et al (2016) present a corpus for prerequisite relations among topics, but this corpus differs in coverage. They used LDA topic modeling to generate a list of 300 topics, while we manually create a list of 200 topics based on criteria described above.…”

“…Recent work has studied the educational aspect of mining text for presenting scientific topics. One goal has been to develop concept maps of topics, graphs showing which topics are prerequisites for learning a given topic (Gordon et al, 2016;Liu et al, 2016;Pan et al, 2017a,b;Liang et al, 2017). Another goal has been to automatically create reading lists for a subject either by building upon concept graphs or through an unstructured approach (Jardine, 2014).…”

Structured Generation of Technical Reading Lists

“…This is strongest in the case of prerequisites (e.g., First-order logic is a prerequisite for understanding Markov logic networks). Gordon et al (2016) propose and evaluate approaches to predict concept dependency relations between LDA topics, and we adopt the average of their best-performing methods:…”

Democratizing data science through data science training

“…For instance, the scraping of textbook tables of contents and course syllabi from the Web is one such means of thematic concepts which occur in a regularized order. For example, given the chapter title “Unsupervised Learning: Clustering and Dimensionality Reduction”, an informed algorithm would select the top five (albeit imperfect) results but with diminishing weights: derived via dimensionality reduction, k-means clustering, fuzzy clustering, machine learning, or sparse dictionary learning, to comprise the semantic relevance vector for the topic set (13). By so doing, we envision an evidence-based means of curriculum development.…”