Research Of Personalized Recommendation Algorithm Based On Reviews And Ratings

As the most successful recommendation technology, the core idea of traditional collaborative filtering algorithm is to calculate user similarity or item similarity from users’ratings and then to predict ratings and recommend items based on similar users’ ratings or similar items’ratings. In real applications, however, there exists a problem of Data Sparsity in which most of users will only give a few ratings. In such a situation, traditional collaborative filtering algorithm cannot produce satisfactory results.Thus, this paper proposes a new topic model based collaborative filtering recommendation framework and two recommendation algorithms, including User-based Collaborative Filtering with Topic Model algorithm (UCFTM) and Item-based Collaborative Filtering with Topic Model algorithm (ICFTM).First of all, each review is processed with topic model to generate corresponding review topic allocations representing a user’s preference for a product’s different features. Next in UCFTM algorithm, we aggregate all topic allocations of reviews given by the same user and calculate user most valued features representing product features that the user most values. In the following, user similarity is calculated based on user most valued features and ratings are predicted from similar users’ratings. In ICFTM algorithm, all topic allocations of reviews given to the same product are also aggregated and item most valued features is calculated representing those most valued features of the product. Then item similarity is calculated based on item most valued features and ratings are predicted from similar items’ratings.Experiments on six real data sets from Amazon indicate that when most of users only give one review and one rating, our algorithms can achieve better results than traditional collaborative filtering algorithms, state-of-the-art topic model based recommendation algorithm and another algorithm dealing with Data Sparsity problem.