Context-Enhanced Citation Sentiment Detection

2012 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 597–601, Montr´eal, Canada, June 3-8, 2012. c⃝2012 Association for Computational Linguistics Context-Enhanced Citation Sentiment Detection Awais Athar University of Cambridge Computer Laboratory 15 JJ Thomson Avenue Cambridge, CB3 0FD, U.K. awais.athar@cl.cam.ac.uk Simone Teufel University of Cambridge Computer Laboratory 15 JJ Thomson Avenue Cambridge, CB3 0FD, U.K. simone.teufel@cl.cam.ac.uk Abstract Sentiment analysis of citations in scientific pa- pers and articles is a new and interesting prob- lem which can open up many exciting new ap- plications in bibliographic search and biblio- metrics. Current work on citation sentiment detection focuses on only the citation sen- tence. In this paper, we address the problem of context-enhanced citation sentiment detec- tion. We present a new citation sentiment cor- pus which has been annotated to take the dom- inant sentiment in the entire citation context into account. We believe that this gold stan- dard is closer to the truth than annotation that looks only at the citation sentence itself. We then explore the effect of context windows of different lengths on the performance of a state- of-the-art citation sentiment detection system when using this context-enhanced gold stan- dard definition. 1 Introduction Sentiment analysis of citations in scientific papers and articles is a new and interesting problem. It can open up many exciting new applications in biblio- graphic search and in bibliometrics, i.e., the auto- matic evaluation of the influence and impact of in- dividuals and journals via citations. Automatic de- tection of citation sentiment can also be used as a first step to scientific summarisation (Abu-Jbara and Radev, 2011). Alternatively, it can help researchers during search, e.g., by identifying problems with a particular approach, or by helping to recognise un- addressed issues and possible gaps in the current re- search. However, there is a problem with the expression of sentiment in scientific text. Conventionally, the writing style in scientific writing is meant to be ob- jective. Any personal bias by authors has to be hedged (Hyland, 1995). Negative sentiment is po- litically particularly dangerous (Ziman, 1968), and some authors have documented the strategy of pref- acing the intended criticism by slightly disingenuous praise (MacRoberts and MacRoberts, 1984). This makes the problem of identifying such opinions par- ticularly challenging. This non-local expression of sentiment has been observed in other genres as well (Wilson et al., 2009; Polanyi and Zaenen, 2006). Figure 1: Example of anaphora in citations A typical case is illustrated in Figure 1. While the first sentence praises some aspects of the cited pa- per, the remaining sentences list its shortcomings. It is clear that criticism is the intended sentiment, but 597 if we define our gold standard only by looking at the citation sentence, we lose a significant amount of sentiment hidden in the text. Given that most ci- tations are neutral (Spiegel-Rosing, 1977; Teufel et al., 2006), this makes it ever more important to re- cover what explicit sentiment there is from the con- text of …

Content truncated to 500 words