Mining User Interests through Internet Review Forum for Building Recommendation System

Research on recommendation system is now getting a lot of attention due to the rapid growth of user generated contents, especially internet review forums. They easily share about their experiences towards some products and services on the review forums. As a result, review forums are overwhelmed with the amount valuable information for predicting user interests. In our work, we present a method to develop a recommendation system leveraging the information mined from review forums. Our method automatically determines user interests by learning from user reviews. Furthermore, we propose the notion of "considered aspects" as the form of user interests, which serve as key information why users are interested in consuming a specific product or service. Several state-of-the-art methods, such as Latent Dirichlet Allocation (LDA), are employed to extract those "considered aspects". Finally, we show that our recommendation system significantly outperforms the baseline system. It is also worth noting that our proposed method is completely unsupervised, domain-independent, and language-independent.