Applying Core Scientific Concepts to Context-Based Citation Recommendation

Applying Core Scientific Concepts to Context-Based Citation Recommendation Daniel Duma1, Maria Liakata2, Amanda Clare3, James Ravenscroft2, Ewan Klein1 University of Edinburgh1, University of Warwick2, University of Aberystwyth3 danielduma@gmail.com, m.liakata@warwick.ed.ac.uk, afc@aber.ac.uk, ravenscroft@papro.org.uk, ewan@inf.ed.ac.uk Abstract The task of recommending relevant scientific literature for a draft academic paper has recently received significant interest. In our effort to ease the discovery of scientific literature and augment scientific writing, we aim to improve the relevance of results based on a shallow semantic analysis of the source document and the potential documents to recommend. We investigate the utility of automatic argumentative and rhetorical annotation of documents for this purpose. Specifically, we integrate automatic Core Scientific Concepts (CoreSC) classification into a prototype context-based citation recommendation system and investigate its usefulness to the task. We frame citation recommendation as an information retrieval task and we use the categories of the annotation schemes to apply different weights to the similarity formula. Our results show interesting and consistent correlations between the type of citation and the type of sentence containing the relevant information. Keywords: context-based, citation recommendation, CoreSC, classification, annotation 1. Introduction Given the need to navigate the ever-increasing volume of scientific literature, the task of Context-Based Citation Rec- ommendation (CBCR) has recently received a lot of inter- est. The task consists in recommending relevant papers to be cited at a specific point in a draft scientific paper, and is universally framed as an information retrieval scenario. However, in order to make these suggestions as useful and relevant as possible, we argue here that we need to apply a measure of understanding to the text of the draft paper. To this end, we investigate the applicability of rhetorical annotation schemes for this task. A number of different schemes for scientific papers have been proposed over the years and several of them have yielded annotated resources, which enable training algorithms for automatic annotation. Like others before, we evaluate our performance at this task by trying to recover the original citations found in papers that have already been published. In an information retrieval scenario like ours, a list of doc- uments ranked by relevance is returned in reply to a query, aiming to satisfy the user’s information need. Our hypoth- esis is that if we can a) classify the information need into a number of discrete categories, and b) classify each sentence in a document according to its function or contribution to it, we could use this segmentation of text to increase the rele- vance of recommendations. 2. CBCR: Previous Work The CBCR task considers that we need to recommend a citation for each citation placeholder: a special token in- serted in the text of a draft paper where the citation should appear. In a standard Information Retrieval (IR) approach, the corpus of potential papers to recommend (the document collection) is indexed for retrieval using a standard vector- space-model approach. Then, for each citation placeholder, the textual context around it (the citation context) is treated as the query, and a similarity measure …

Content truncated to 500 words