The Haves and the Have-Nots: Leveraging Unlabelled Corpora for Sentiment Analysis

Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, pages 412–422, Sofia, Bulgaria, August 4-9 2013. c⃝2013 Association for Computational Linguistics The Haves and the Have-Nots: Leveraging Unlabelled Corpora for Sentiment Analysis Kashyap Popat2 Balamurali A R1,2,3 Pushpak Bhattacharyya2 Gholamreza Haffari3 1IITB-Monash Research Academy, IIT Bombay 3Monash University 2Dept. of Computer Science and Engineering, IIT Bombay Australia {kashyap,balamurali,pb}@cse.iitb.ac.in reza@monash.edu Abstract Expensive feature engineering based on WordNet senses has been shown be useful for document level sentimentto classification. A plausible reason for such a performance improvement is the reduction in data sparsity. However, such a reduction could be achieved with a lesser effort through the means of syntagma based word clustering. In this paper, the problem of data sparsity in sentiment analysis, both monolingual and cross-lingual, is addressed through the means of clustering. Experiments show that cluster based data sparsity reduction leads to performance better than sense based classification for sentiment analysis at document level. Similar idea is applied to Cross Lingual Sentiment Analysis (CLSA), and it is shown that reduction in data sparsity (after translation or bilingual-mapping) produces accuracy higher than Machine Translation based CLSA and sense based CLSA. 1 Introduction Data sparsity is the bane of Natural Language Processing (NLP) (Xue et al., 2005; Minkov et al., 2007). Language units encountered in the test data but absent in the training data severely degrade the performance of an NLP task. NLP applications innovatively handle data sparsity through various means. A special, but very common kind of data sparsity viz., word sparsity, can be addressed in one of the two obvious ways: 1) sparsity reduction through paradigmatically related words or 2) sparsity reduction through syntagmatically related words. Paradigmatic analysis of text is the analysis of concepts embedded in the text (Cruse, 1986; Chandler, 2012). WordNet is a byproduct of such an analysis. In WordNet, paradigms are manually generated based on the principles of lexical and semantic relationship among words (Fellbaum, 1998). WordNets are primarily used to address the problem of word sense disambiguation. However, at present there are many NLP applications which use WordNet. One such application is Sentiment Analysis (SA) (Pang and Lee, 2002). Recent research has shown that word sense based semantic features can improve the performance of SA systems (Rentoumi et al., 2009; Tamara et al., 2010; Balamurali et al., 2011) compared to word based features. Syntagmatic analysis of text concentrates on the surface properties of the text. Compared to paradigmatic property extraction, syntagmatic processing is relatively light weight. One of the obvious syntagmas is words, and words are grouped into equivalence classes or clusters, thus reducing the model parameters of a statistical NLP system (Brown et al., 1992). When used as an additional feature with word based language models, it has been shown to improve the system performance viz., machine translation (Uszkoreit and Brants, 2008; Stymne, 2012), speech recognition (Martin et al., 1995; Samuelsson and Reichl, 1999), dependency parsing (Koo et al., 2008; Haffari et al., 2011; Zhang and Nivre, 2011; Tratz and Hovy, …

Content truncated to 500 words