Understanding Online Consumer Review Opinions with Sentiment Analysis using Machine Learning
With the advent of Web 2.0 technologies, the Web has evolved to become a popular channel of communication and interaction between Web users and online consumers. Social media, unlike traditional media, have rich but unorganized content contributed by users, often in fragmented and sparse fashion. Users usually spend a lot of their time filtering useless information and yet are not able to capture the essence. In this study, we focus on user-contributed reviews of products, which many online consumers use to support their purchase decisions by identifying products that best fit their preferences. In the recent years, sentiment classification and analysis of online consumer reviews has drawn significant research attention. Most existing techniques rely on natural language processing tools to parse and analyze sentences in a review, yet they offer poor accuracy, because the writing in online reviews tends to be less formal than writing in news or journal articles. Many opinion sentences contain grammatical errors and unknown terms that do not exist in dictionaries. Therefore, this study proposes two supervised learning techniques (class association rules and naïve Bayes classifier) to classify opinion sentences into appropriate product feature classes and produce a summary of consumer reviews. An empirical evaluation that compares the performance of the class association rules technique and the naïve Bayes classifier for sentiment analysis shows that our proposed techniques achieve more than 70% of the macro and micro F-measures.