ECSCite : a system for the automatic extraction, clustering and summarization of citation contexts

Abstract

Academic contributions to knowledge are often inspired by the ideas and results published in contributions made by others in the past. The origins of such inspiration are typically acknowledged in academic texts by means of citations or references that explicitly attribute these ideas to their sources. Academic texts are thereby linked together by citations made; each text is inherently connected to the other academic texts that it cites. Collections of domain-related academic contributions can hence to exhibits a web-like, networked structure. Researchers traverse this structure to trace the genealogy of ideas that are of interest to them using the contextual information provided with citations made in reference to previous works to guide them. The contextual information provided with a citation, which is often referred to as a citation context, is a textual description that contains a key fact, from an academic author's perspective, about the cited document. The set of all citation contexts made in reference to a particular academic paper can thus be seen to be representative of the salient facts of the contribution that were deemed relevant by the academic community. Such knowledge is valuable to researcher for a variety of reasons, such as: in assessing the relevance of the cited paper; in acquiring feedback about ideas in the paper; in discovering citing papers of interest; and many more. Researchers use this information all the time but they have to manually construct this set, or parts of it, by inspecting each paper for relevant citations and analyzing the context of each one. Such analyses are often quite time consuming, especially for high-impact papers that are cited by many others. We present ECSCite, a system for the automatic Extraction, Clustering and Summarization of Citation contexts that automates the process described above to present to its users summarized overviews of the different ways an academic document of interest was cited by others. ECSCite thus facilitates the analysis of citation contexts, sparing its users of collecting and organizing data themselves. It does this by making use of an existing citation indexing engine to locate other academic documents that cite a given document and constructing the set of all citation contexts that reference the contribution by extracting them from each of the citing documents found. It clusters the set into semantically related groups of contexts that highlight the same key facts c:ibout the cited contribution, automatically generates a summary of each of the clustered groups, and finally presents these results for perusal by its users.