Towards a Generic and Flexible Citation Classifier Based on a Faceted Classification Scheme

Proceedings of COLING 2012: Technical Papers, pages 1343–1358, COLING 2012, Mumbai, December 2012. Towards a Generic and Flexible Citation Classifier Based on a Faceted Classification Scheme Charles Jochim Hinrich Schütze Institute for Natural Language Processing, University of Stuttgart charles.jochim@ims.uni-stuttgart.de ABSTRACT Citations are a valuable resource for characterizing scientific publications that has already been used in applications such as summarization and information retrieval. These applications could be even better served by expanding citation information. We aim to achieve this by extracting and classifying citation information from the text, so that subsequent applications may make use of it. We make three contributions to the advancement of fine-grained citation classification. First, our work uses a standard classification scheme for citations that was developed independently of automatic classification and therefore is not bound to any particular citation application. Second, to address the lack of available annotated corpora and reproducible results for citation classification, we are making available a manually-annotated corpus as a benchmark for further citation classification research. Third, we introduce new features designed for citation classification and compare them experimentally with previously proposed citation features, showing that these new features improve classification accuracy. KEYWORDS: citation classification, feature extraction. 1343 1 Introduction Citations are a valuable resource for characterizing scientific publications and their links to each other. They have been exploited for a number of natural language processing (NLP) and information retrieval (IR) applications, including summarization (Qazvinian and Radev, 2008; Qazvinian et al., 2010)[CJPF]1, improved indexing and retrieval (Ritchie et al., 2006)[CJPF], and building integrated research databases (Nanba et al., 2004)[CJPF]. Bibliometric measures that quantify the impact of publications (e.g., Moed, 2005)[CJPF]are also based on citations. Most of this work does not differentiate between uses of citations, e.g., whether a citation is more or less important to the paper or whether the paper’s authors support or refute the claims made in the cited work. However, recently a number of research groups have attempted to classify citations with respect to dimensions like importance and relation to cited work (Teufel et al., 2006b; Dong and Schäfer, 2011; Sugiyama et al., 2010; Abu-Jbara and Radev, 2012)[CEPF]. By adding such fine-grained information to individual citations, the various applications of citation analysis can be better served; e.g., citations that are foundational to a paper may constitute better summary sentences for the cited paper. Thus, there are clear potential benefits to fine-grained citation analysis; and a number of case studies have been published that demonstrate this potential (Nanba et al., 2004; Teufel et al., 2006b)[CEPF]. However, fine-grained citation analysis is currently not widely used in applications that access and analyze the scientific literature. In this paper, we identify a number of potential reasons for this state of affairs and propose solutions. The first problem with current fine-grained citation analysis is that prior work has tended to develop custom classification schemes for a particular application. This means that the development cycle for a citation classifier must be started from scratch for each new application. In contrast to this prior work, we …

Content truncated to 500 words