Sentiment Classification with Graph Co-Regularization

Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers, pages 1331–1340, Dublin, Ireland, August 23-29 2014. Sentiment Classification with Graph Co-Regularization Guangyou Zhou, Jun Zhao, and Daojian Zeng National Laboratory of Pattern Recognition Institute of Automation, Chinese Academy of Sciences 95 Zhongguancun East Road, Beijing 100190, China {gyzhou,jzhao,djzeng}@nlpr.ia.ac.cn Abstract Sentiment classification aims to automatically predict sentiment polarity (e.g., positive or neg- ative) of user-generated sentiment data (e.g., reviews, blogs). To obtain sentiment classifica- tion with high accuracy, supervised techniques require a large amount of manually labeled data. The labeling work can be time-consuming and expensive, which makes unsupervised (or semi- supervised) sentiment analysis essential for this application. In this paper, we propose a novel algorithm, called graph co-regularized non-negative matrix tri-factorization (GNMTF), from the geometric perspective. GNMTF assumes that if two words (or documents) are sufficiently close to each other, they tend to share the same sentiment polarity. To achieve this, we encode the geometric information by constructing the nearest neighbor graphs, in conjunction with a non- negative matrix tri-factorization framework. We derive an efficient algorithm for learning the factorization, analyze its complexity, and provide proof of convergence. Our empirical study on two open data sets validates that GNMTF can consistently improve the sentiment classification accuracy in comparison to the state-of-the-art methods. 1 Introduction Recently, sentiment classification has gained a wide interest in natural language processing (NLP) com- munity. Methods for automatically classifying sentiments expressed in products and movie reviews can roughly be divided into supervised and unsupervised (or semi-supervised) sentiment analysis. Super- vised techniques have been proved promising and widely used in sentiment classification (Pang et al., 2002; Pang and Lee, 2008; Liu, 2012). However, the performance of these methods relies on manually labeled training data. In some cases, the labeling work may be time-consuming and expensive. This motivates the problem of learning robust sentiment classification via unsupervised (or semi-supervised) paradigm. A traditional way to perform unsupervised sentiment analysis is the lexicon-based method (Turney, 2002; Taboada et al., 2011). Lexicon-based methods employ a sentiment lexicon to determine overall sentiment orientation of a document. However, it is difficult to define a universally optimal sentiment lexicon to cover all words from different domains (Lu et al., 2011a). Besides, most semi-automated lexicon-based methods yield unsatisfactory lexicons, with either high coverage and low precision or vice versa (Ng et al., 2006). Thus it is challenging for lexicon-based methods to accurately identify the overall sentiment polarity of users generated sentiment data. Recently, Li et al. (2009) proposed a constrained non-negative matrix tri-factorization (CNMTF) approach to sentiment classification, with a domain-independent sentiment lexicon as prior knowledge. Experimental results show that CNMTF achieves state-of-the-art performance. From the geometric perspective, the data points (words or documents) may be sampled from a distribu- tion supported by a low-dimensional manifold embedded in a high-dimensional space (Cai et al., 2011). This geometric structure, meaning that two words (or documents) sufficiently close to each other tend to share the same sentiment polarity, should be preserved during the …

Content truncated to 500 words