Topic Identification for Fine-Grained Opinion Analysis

Proceedings of the 22nd International Conference on Computational Linguistics (Coling 2008), pages 817–824 Manchester, August 2008 Topic Identification for Fine-Grained Opinion Analysis Veselin Stoyanov and Claire Cardie Department of Computer Science Cornell University {stoyanov,cardie}@cs.cornell.edu Abstract Within the area of general-purpose fine- grained subjectivity analysis, opinion topic identification has, to date, received little attention due to both the difficulty of the task and the lack of appropriately anno- tated resources. In this paper, we pro- vide an operational definition of opinion topic and present an algorithm for opinion topic identification that, following our new definition, treats the task as a problem in topic coreference resolution. We develop a methodology for the manual annotation of opinion topics and use it to annotate topic information for a portion of an existing general-purpose opinion corpus. In exper- iments using the corpus, our topic identi- fication approach statistically significantly outperforms several non-trivial baselines according to three evaluation measures. 1 Introduction Subjectivity analysis is concerned with extract- ing information about attitudes, beliefs, emotions, opinions, evaluations, sentiment and other private states expressed in texts. In contrast to the prob- lem of identifying subjectivity or sentiment at the document level (e.g. Pang et al. (2002), Turney (2002)), we are interested in fine-grained subjec- tivity analysis, which is concerned with subjec- tivity at the phrase or clause level. We expect fine-grained subjectivity analysis to be useful for question-answering, summarization, information extraction and search engine support for queries of c⃝ 2008. Licensed under the Creative Commons Attribution-Noncommercial-Share Alike 3.0 Unported li- cense (http://creativecommons.org/licenses/by-nc-sa/3.0/). Some rights reserved. the form “How/what does entity X feel/think about topic Y?”, for which document-level opinion anal- ysis methods can be problematic. Fine-grained subjectivity analyses typically identify SUBJECTIVE EXPRESSIONS in context, charac- terize their POLARITY (e.g. positive, neutral or neg- ative) and INTENSITY (e.g. weak, medium, strong, extreme), and identify the associated SOURCE, or OPINION HOLDER, as well as the TOPIC, or TARGET, of the opinion. While substantial progress has been made in automating some of these tasks, opinion topic identification has received by far the least at- tention due to both the difficulty of the task and the lack of appropriately annotated resources.1 This paper addresses the problem of topic iden- tification for fine-grained opinion analysis of gen- eral text.2 We begin by providing a new, opera- tional definition of opinion topic in which the topic of an opinion depends on the context in which its associated opinion expression occurs. We also present a novel method for general-purpose opin- ion topic identification that, following our new def- inition, treats the problem as an exercise in topic coreference resolution. We evaluate the approach using the existing MPQA corpus (Wiebe et al., 2005), which we extend with manual annotations that encode topic information (and refer to here- after as the MPQATOPIC corpus). Inter-annotator agreement results for the manual annotations are reasonably strong across a num- ber of metrics and the results of experiments that evaluate our topic identification method in the con- text of fine-grained opinion analysis …

Content truncated to 500 words