Unsupervised Discovery of Discourse Relations for Eliminating Intra-sentence Polarity Ambiguities

Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing, pages 162–171, Edinburgh, Scotland, UK, July 27–31, 2011. c⃝2011 Association for Computational Linguistics Unsupervised Discovery of Discourse Relations for Eliminating Intra-sentence Polarity Ambiguities Lanjun Zhou, Binyang Li, Wei Gao, Zhongyu Wei, Kam-Fai Wong Department of Systems Engineering and Engineering Management The Chinese University of Hong Kong Shatin, NT, Hong Kong, China Key Laboratory of High Confidence Software Technologies Ministry of Education, China {ljzhou, byli, wgao, zywei, kfwong}@se.cuhk.edu.hk Abstract Polarity classification of opinionated sen- tences with both positive and negative senti- ments1 is a key challenge in sentiment anal- ysis. This paper presents a novel unsuper- vised method for discovering intra-sentence level discourse relations for eliminating polar- ity ambiguities. Firstly, a discourse scheme with discourse constraints on polarity was de- fined empirically based on Rhetorical Struc- ture Theory (RST). Then, a small set of cue- phrase-based patterns were utilized to collect a large number of discourse instances which were later converted to semantic sequential representations (SSRs). Finally, an unsuper- vised method was adopted to generate, weigh and filter new SSRs without cue phrases for recognizing discourse relations. Experimen- tal results showed that the proposed methods not only effectively recognized the defined discourse relations but also achieved signifi- cant improvement by integrating discourse in- formation in sentence-level polarity classifica- tion. 1 Introduction As an important task of sentiment analysis, polar- ity classification is critically affected by discourse structure (Polanyi and Zaenen, 2006). Previous re- search developed discourse schema (Asher et al., 2008) (Somasundaran et al., 2008) and proved that the utilization of discourse relations could improve the performance of polarity classification on dia- logues (Somasundaran et al., 2009). However, cur- 1Defined as ambiguous sentences in this paper rent state-of-the-art methods for sentence-level po- larity classification are facing difficulties in ascer- taining the polarity of some sentences. For example: (a) [Although Fujimori was criticized by the international community]，[he was loved by the domestic population]， [because people hated the corrupted ruling class]. (儘管 國際間對藤森口誅筆伐，他在國內一直深受百姓愛 戴，原因是百姓對腐化的統治階級早就深惡痛絕。) Example (a) is a positive sentence holding a Con- trast relation between first two segments and a Cause relation between last two segments. The po- larity of "criticized", "hated" and "corrupted" are rec- ognized as negative expressions while "loved" is rec- ognized as a positive expression. Example (a) is dif- ficult for existing polarity classification methods for two reasons: (1) the number of positive expressions is less than negative expressions; (2) the importance of each sentiment expression is unknown. However, consider Figure 1, if we know that the polarity of the first two segments holding a Contrast relation is determined by the nucleus (Mann and Thompson, 1988) segment and the polarity of the last two seg- ments holding a Cause relation is also determined by the nucleus segment, the polarity of the sentence will be determined by the polarity of "[he...population]". Thus, the polarity of Example (a) is positive. Statistics showed that 43% of the opinionated sentences in NTCIR2 MOAT (Multilingual Opinion Analysis Task) Chinese corpus3 are …

Content truncated to 500 words