ISSN: 2229-6956 (ONLINE) ICTACT JOURNAL ON SOFT COMPUTING, JULY 2017, VOLUME: 07, ISSUE: 04 DOI: 10.21917/ijsc.2017.0211 1517 A HYBRID APPROACH FOR POLARITY SHIFT DETECTION Michele Mistry 1 and Prem Balani 2 Department of Information Technology, G.H. Patel College of Engineering and Technology, India Abstract Now-a-days sentiment analysis has become a hot research area. With the increasing use of internet, people express their views by using social media, blogs, etc. So there is a dire need to analyze people’s opinions. Sentiment classification is the main task of sentiment analysis. But while classifying sentiments, the problem of polarity shift occurs. Polarity shift is considered as a very crucial problem. Polarity shift changes a text from positive to negative and vice versa. In this paper, a hybrid approach is proposed for polarity shift detection of negation (explicit and implicit) and contrast. The hybrid approach consists of a rule-based approach for detecting explicit negation and contrast and a lexicon called SentiWordNet for detecting implicit negation. The proposed approach outperforms its baselines. Keywords Sentiment Analysis, Sentiment Classification, Polarity Shift, Natural Language Processing, Lexicon 1. INTRODUCTION In recent years the use of internet and e-commerce has increased. More and more products are sold on the Web, and the people buying the products write reviews on it. The data that Web contains is in the form of product reviews, news, blogs, internet forums etc. The volume of online reviews available on the Internet is growing day by day. As a result of this growth, sentiment analysis has become a hot field in the area of natural language processing. Natural language processing (NLP) is actually a theory-motivated range of computational techniques that are used for the automatic analysis and representation of human language [34].The term sentiment analysis first appeared in the work of Nasukawa and Yi [30]. The term opinion mining first appeared in the work of Dave, Lawrence and Pennock [29]. But the research on sentiments and opinions appeared much earlier [2], [8], [9], [31], [32], [33]. In literature, subjectivity and emotion are closely related to sentiment and opinion. An objective sentence is the sentence that represents some factual information about the world. Whereas, a subjective sentence expresses some feelings, views or beliefs [1]. Objective Sentence: “iPhone is an Apple product.” Subjective Sentence: “I like iPhone.” Sentiments consist of feelings, thoughts and emotions of an individual for a particular event or topic [11]. According to Bing Liu [1], “Sentiment analysis, also called opinion mining, is the field of study that analyzes people’s opinions, sentiments, evaluations, appraisals, attitudes, and emotions towards entities such as products, services, organizations, individuals, issues, events, topics, and their attributes. And the most important indicators of sentiments are sentiment words or opinion words.” The basic task in sentiment analysis is sentiment classification. Sentiment classification is considered to be a topic that is studied very extensively in recent times [7]. Sentiment classification is used to classify the text as positive, negative or neutral [1]. While doing sentiment classification, many times there occurs a problem called the polarity shift problem. Detecting the polarity shifts is a very important task for sentiment classification. Polarity shift is considered to be a linguistic phenomenon. In polarity shift, the polarity of a particular review changes from positive to negative or vice versa [3]. While doing sentiment classification, if the Bag-of-words (BOW) model is used then it will not regard any grammar, as a result of that the syntactic structure of the sentence may be disturbed and that’ll cause problems like polarity shift detection, anaphora resolution, etc. [6]. The words that change the polarity of the text are called polarity shifters. They are also called “valence shifter” [5] and “Sentiment shifter” [1]. There are many kinds of polarity shifters like explicit and implicit negation, contrast, likelihood, counter factual, etc. By detecting the polarity shifters, one can know the reason that has changed the polarity of the text. For e.g. “I am not satisfied with the working of this juicer.” In the above review, “not” is the polarity shifter that is changing the polarity of the review to negative. The linguistic phenomenon in which the sentimental orientation of the whole text is changed from negative to positive or vice versa; is called polarity shifting [3], [4]. Polarity shift is caused by polarity shifters, e.g., contrast, negation, likelihood, etc. [6]. Negation is considered as the most common type of polarity shift. Negation is of two types: explicit negation and implicit negation. Explicit negators are words like “no”, “not” etc. Implicit negators are words like “avoid”, “deny” etc. The various approaches that are used for polarity shift detection are: machine learning approaches [2], [8], lexicon-based approaches [11], [13], [17] and hybrid approaches [18]. These are widely used along with some rule based approaches and statistical approaches. This paper presents a hybrid approach for polarity shift detection. A rule-based method for detecting the explicit negators and contrasts is used. An updated negation list using SentiStrength lexicon has been used. SentiWordNet has been used to detect implicit negators and compare the obtained results with baseline approaches. The rest of the paper is organized as follows: Section 2 contains the related work. Section 3 contains the proposed work, section 4 contains the experimental results and Section 5 contains the conclusion. 2. RELATED WORK Sentiment analysis is a growing research field [3]. According to Bing Liu [1], “Sentiment analysis is considered as a highly restricted NLP problem. It is so because there is no need for the system to fully understand the semantics of each sentence or document but it only needs to understand some aspects of it, i.e., positive or negative sentiments and their target entities or topics.”