Feature Selection Methods in Sentiment Analysis and Sentiment Classification of Amazon Product Reviews

摘要

Sentiment Analysis or Opinion Mining is a nascent field of data mining, which is expanding and much research work is being done in this field. Opinion Mining mines people’s opinion towards a topic. Opinion mining’s main objective is to extract opinion or views of a person for a particular topic or subject. Mainly Opinion Mining classifies the given review as positive, neutral or negative. Opinion Mining has accomplished much focus nowadays due to availability of vast amount of opinion rich web resources such as online product reviews, blogs, social networking sites etc. As the use of ecommerce websites are increasing profusely and people are opting for online shopping there is vast amount of data generated which can be useful for Opinion Mining. In this paper, different feature extraction or selection techniques for opinion mining are performed. Work is carried out in different steps. First step is the data collection step in which amazon dataset is used. Second is the preprocessing step which is used for the removal of stop words and special characters. In the third step, feature selection or extraction techniques like phrase level, single word and multiword are applied over the amazon dataset. The fourth step is used to generate the vector of the extracted features. In the final step, Na?ve Bayes classifier is applied to classify the reviews. Step one to four is used for training the system and last step is used for testing. In the paper Supervised learning method is used for classification of reviews.