Web Personalization Using Opinion Mining

Abstract

Referring to reviews, checking online comments and, visiting different websites before buying any product is a call of the day. People go through reviews even before they go for dinner in a restaurant. Mundane activities like purchasing daily wear or grocery or even vegetables are heavily dominated by recommendations from friends, family members, colleagues etc. Thus online reviews are an excellent source of information both for users and organizations alike. In this thesis, a hybrid model, named aspect and context-based latent factor model (ACMF) is proposed to predict user rating on an item based on star ratings provided by users, feature-opinion information, and context information. ACMF mainly consists of three phases: The first phase extracts spam reviews and discards them, the second phase extracts features and opinions from written reviews and eventually calculates the polarity score of opinions. In the last phase, ratings, reviews, and context information are aggregated to predict the unknown rating of a user for better recommendations and further improve user experience and address the sparsity problem. The proposed model is tested on ratings and reviews downloaded from the Amazon website. Experiment results show root mean squared error(RMSE) of ACMF has been achieved significantly less than other relevant methods that indicate that incorporating sentiment and context aspects in the latent factor model can aid in improving the recommendation performance. newline