SENTIMENT ANALYSIS AND CLASSIFICATION USING LEXICON-BASED APPROACH AND ADDRESSING POLARITY SHIFT PROBLEM

• Published in: Journal of Theoretical and Applied Information Technology 2016-08, Vol.90 (1), p.118-118
• Main Authors: Kolekar, Nilam V, Rao, Gauri, Dey, Sumita, Mane, Madhavi, Jadhav, Veena, Patil, Shruti
• Format: Article
• Language: English
• Subjects: Classification; Construction; Data mining; Markets; Orientation; Polarity; Sentences; Texts
• ISSN: 1817-3195

As we know that in recent years e-commerce has been growing, so volume of online reviews on the web is also increasing for different sides due to which we can understand that the particular product or things are good for use or not and their current status in market. In Natural Language Processing (NLP) and text mining, different models and methods are useful for text representation and categorization purposes. Bag-Of-Word (BOW) model is one such model used to model the text. But polarity shift problem is a major factor in Bag-Of-Word model which can effect on classification performance of Sentiment Analysis. In our methodology, we address the polarity shift problem by detecting, removing and modifying negation from extracted review to identify where the sentiment orientation is actually changing in given review. Our main idea is Sentiment analysis and classification which is based on machine learning approach using Lexicon based antonym dictionary. We build system for Sentence-level sentiment classification. We first extract product reviews from one of the customized shopping portal. When extracted reviews are simple sentences then system is trained to directly find its opinion target, and classify it according to its sentiment polarities i.e. is positive, negative and neutral class labels. When extracted reviews are compound and complex sentences then we first split it into subsentences and build a model based on some rules to detect, remove and modify polarity shift in contrast of negation to identify where the sentiment orientation is changing in compound or complex sentences. After that, we classify review according to its polarity and determine the targets of opinion given in review. Furthermore, we extend our system for opinion summarization based on opinion features or aspects and graphically represent overall summary of Positive, Negative and Neutral sentiments of customer for each product.