Learning to Shift the Polarity of Words for Sentiment Classification

Learning to Shift the Polarity of Words for Sentiment Classification Daisuke Ikeda† Hiroya Takamura‡ Lev-Arie Ratinov†† Manabu Okumura‡ †Department of Computational Intelligence and Systems Science, Tokyo Institute of Technology ikeda@lr.pi.titech.ac.jp ††Department of Computer Science, University of Illinois at Urbana-Champaign ratinov2@uiuc.edu ‡Precision and Intelligence Laboratory, Tokyo Institute of Technology {takamura,oku}@pi.titech.ac.jp Abstract We propose a machine learning based method of sentiment classification of sen- tences using word-level polarity. The polari- ties of words in a sentence are not always the same as that of the sentence, because there can be polarity-shifters such as negation ex- pressions. The proposed method models the polarity-shifters. Our model can be trained in two different ways: word-wise and sentence-wise learning. In sentence-wise learning, the model can be trained so that the prediction of sentence polarities should be accurate. The model can also be combined with features used in previous work such as bag-of-words and n-grams. We empiri- cally show that our method almost always improves the performance of sentiment clas- sification of sentences especially when we have only small amount of training data. 1 Introduction Due to the recent popularity of the internet, individ- uals have been able to provide various information to the public easily and actively (e.g., by weblogs or online bulletin boards). The information often in- cludes opinions or sentiments on a variety of things such as new products. A huge amount of work has been devoted to analysis of the information, which is called sentiment analysis. The sentiment analysis has been done at different levels including words, sentences, and documents. Among them, we focus on the sentiment classification of sentences, the task to classify sentences into “positive” or “negative”, because this task is fundamental and has a wide ap- plicability in sentiment analysis. For example, we can retrieve individuals’ opinions that are related to a product and can find whether they have the positive attitude to the product. There has been much work on the identification of sentiment polarity of words. For instance, “beauti- ful” is positively oriented, while “dirty” is negatively oriented. We use the term sentiment words to refer to those words that are listed in a predefined polar- ity dictionary. Sentiment words are a basic resource for sentiment analysis and thus believed to have a great potential for applications. However, it is still an open problem how we can effectively use sen- timent words to improve performance of sentiment classification of sentences or documents. The simplest way for that purpose would be the majority voting by the number of positive words and the number of negative words in the given sentence. However, the polarities of words in a sentence are not always the same as that of the sentence, be- cause there can be polarity-shifters such as nega- tion expressions. This inconsistency of word-level polarity and sentence-level polarity often causes er- rors in classification by the simple majority voting method. A manual list of polarity-shifters, which are the words that can shift the sentiment polarity of another word (e.g., negations), …

Content truncated to 500 words