Sentiment Lexicon Interpolation and Polarity Estimation of Objective and Out-Of-Vocabulary Words to Improve Sentiment Classification on Microblogging

PACLIC 28 !204 Sentiment Lexicon Interpolation and Polarity Estimation of Objective and Out-Of-Vocabulary Words to Improve Sentiment Classification on Microblogging Yongyos Kaewpitakkun, Kiyoaki Shirai, Masnizah Mohd Japan Advanced Institute of Science and Technology 1-1, Asahidai, Nomi City, Ishikawa, Japan 923-1292 Email: {s1320203,kshirai,masnizah}@jaist.ac.jp Abstract Sentiment analysis has become an important classification task because a large amount of user-generated content is published over the Internet. Sentiment lexicons have been used successfully to classify the sentiment of user review datasets. More recently, microblog- ging services such as Twitter have become a popular data source in the domain of senti- ment analysis. However, analyzing sentiments on tweets is still difficult because tweets are very short and contain slang, informal expres- sions, emoticons, mistyping and many words not found in a dictionary. In addition, more than 90 percent of the words in public senti- ment lexicons, such as SentiWordNet, are ob- jective words, which are often considered less important in a classification module. In this paper, we introduce a hybrid approach that in- corporates sentiment lexicons into a machine learning approach to improve sentiment clas- sification in tweets. We automatically con- struct an Add-on lexicon that compiles the po- larity scores of objective words and out-of- vocabulary (OOV) words from tweet corpora. We also introduce a novel feature weight- ing method by interpolating sentiment lexi- con score into uni-gram vectors in the Support Vector Machine (SVM). Results of our exper- iment show that our method is effective and significantly improves the sentiment classifi- cation accuracy compared to a baseline uni- gram model. 1 Introduction Sentiment analysis and opinion mining is the field of study that analyzes people’s opinions, sentiments, evaluations, attitudes and emotions from written lan- guage (Liu, 2010). Recently, Twitter has become an important resource for sentiment analysis. Peo- ple express their opinions and feelings using Twit- ter and these data can be grabbed publicly through Twitter API. There are two main approaches to senti- ment analysis: lexicon-based and machine learning- based techniques. Several researchers have com- bined these two techniques (Kumar et al., 2012; Mudinas et al., 2012; Saif et al., 2012; Fang et al., 2011; Hung et al., 2013). This study adopts a simi- lar approach; we seek to combine the prior polarity knowledge from the lexicon-based method and the powerful classification algorithm from the machine learning-based method. Two main motivations of this approach are discussed below. The initial motivation is to revise the polarity of objective and out-of-vocabulary words in the public sentiment lexicon to improve Twitter sentiment clas- sification. In the lexicon-based approach, sentiment classification is done by comparing the group of pos- itive and negative words looked up from the public lexicon. For example, if the document contains more positive words than negative words, it will be clas- sified as positive. Several public lexical resources such as ANEW1, OpinionFinder2, SentiStrength3, SentiWordNet4 and SenticNet5 lexicon are available for this type of analysis. SentiWordNet or “SWN” (Esuli et al., 2010) has become one of the most fa- 1http://neuro.imm.dtu.dk/wiki/A new ANEW/ 2http://mpqa.cs.pitt.edu/opinionfinder/ …

Content truncated to 500 words