Research On Collaborative Filtering Recommendation Algorithm Based On Reviews And Ratings

Collaborative filtering algorithm is one of the commonly used and most successful recommendation technology, which has been successfully applied to many online applications. But the traditional collaborative filtering algorithms have the problem called data sparsity problem, and when the user rating data is scarce or missing, recommender system will be difficult to make an accurate rating predict and item recommendation. In recent years, the researchers are paying more attention to the research on review analysis and text mining work. Compared with the rating data, the user review often contain more abundant and valuable information resources. In order to overcome the sparsity problem of traditional collaborative filtering algorithm, by combining the characteristics of user ratings and user reviews, this paper proposed User-based Collaborative Filtering with Reviews and Ratings, and Item-based Collaborative Filtering with Reviews and Ratings.The main work includes:(1) Introduce the research background of this paper and the development situation of recommendation technology. Analyze current major recommendation algorithm, and briefly discuss about the evaluation method of recommendation system.(2) Analyze the features and elements of user review, and introduce the LDA topic model. Detailed introduce the user-based collaborative filtering algorithm and the item-based collaborative filtering algorithm. Then this paper proposes review topic distribution, review attitude, improved user preference, and the improved item feature.(3) According to the characteristics of user reviews and utilize the basic idea of traditional collaborative filtering algorithm, two new collaborative filtering algorithms are proposed in this paper: user-based collaborative filtering with reviews and ratings(UCFRR), and item-based collaborative filtering with reviews and ratings(ICFRR). The algorithms utilized topic model to generate review topics distribution, and utilized ratings to generate review attitude. Then used the review attitude and review topics to establish the more accurate user preference model and item feature model, and finally predict ratings and make recommendations.(4) Use the electronic device reviews dataset of Amazon to evaluate the UCFRR and ICFRR, and compare with other existing algorithms.The experimental results show that, comparing with traditional collaborative filtering algorithm and other existing algorithms which based on topic model, UCFRR and ICFRR effectively improve the rating prediction accuracy and the quality of recommendation.