Sentiment Classification and Polarity Shifting

Proceedings of the 23rd International Conference on Computational Linguistics (Coling 2010), pages 635–643, Beijing, August 2010 Sentiment Classification and Polarity Shifting Shoushan Li†‡ Sophia Yat Mei Lee† Ying Chen† Chu-Ren Huang† Guodong Zhou‡ †Department of CBS The Hong Kong Polytechnic University {shoushan.li, sophiaym, chenying3176, churenhuang} @gmail.com ‡ Natural Language Processing Lab School of Computer Science and Technology Soochow University gdzhou@suda.edu.cn Abstract Polarity shifting marked by various linguistic structures has been a challenge to automatic sentiment classification. In this paper, we propose a machine learning approach to incorporate polarity shifting information into a document-level sentiment classification system. First, a feature selection method is adopted to automatically generate the training data for a binary classifier on polarity shifting detection of sentences. Then, by using the obtained binary classifier, each document in the original polarity classification training data is split into two partitions, polarity-shifted and polarity-unshifted, which are used to train two base classifiers respectively for further classifier combination. The experimental results across four different domains demonstrate the effectiveness of our approach. 1 Introduction Sentiment classification is a special task of text classification whose objective is to classify a text according to the sentimental polarities of opinions it contains (Pang et al., 2002), e.g., favorable or unfavorable, positive or negative. This task has received considerable interests in the computational linguistic community due to its potential applications. In the literature, machine learning approaches have dominated the research in sentiment classification and achieved the state-of-the-art performance (e.g., Kennedy and Inkpen, 2006; Pang et al., 2002). In a typical machine learning approach, a document (text) is modeled as a bag-of-words, i.e. a set of content words without any word order or syntactic relation information. In other words, the underlying assumption is that the sentimental orientation of the whole text depends on the sum of the sentimental polarities of content words. Although this assumption is reasonable and has led to initial success, it is linguistically unsound since many function words and constructions can shift the sentimental polarities of a text. For example, in the sentence ‘The chair is not comfortable’, the polarity of the word ‘comfortable’ is positive while the polarity of the whole sentence is reversed because of the negation word ‘not’. Therefore, the overall sentiment of a document is not necessarily the sum of the content parts (Turney, 2002). This phenomenon is one main reason why machine learning approaches fail under some circumstances. As a typical case of polarity shifting, negation has been paid close attention and widely studied in the literature (Na et al., 2004; Wilson et al., 2009; Kennedy and Inkpen, 2006). Generally, there are two steps to incorporate negation information into a system: negation detection and negation classification. For negation detection, some negation trigger words, such as ‘no’, ‘not’, and ‘never’, are usually applied to recognize negation phrases or sentences. As for negation classification, one way to import negation information is to directly reverse the polarity of the words which contain negation trigger words as far as term-counting approaches are considered (Kennedy …

Content truncated to 500 words