Incorporating user rating credibility in recommender systems

There has been a lot of research efforts aimed at improving the recommendation accuracy with Collaborative Filtering (CF); yet there is still a lack of investigation into the integration of CF algorithms with the analysis of users' rating behaviors. Considering that by incorporating the rating credibility, the impact of the ratings given by neighbors with low credibility should be decreased. In this work, we develop an integrated solution for CF recommendation by incorporating the credibility of users' ratings, demographic information of the people, and ontological semantics of items. The demographic information of users and ontological semantics of items are used in the similarity measurement of users/items to alleviate the issues of sparsity and cold-start in CF algorithms. To our knowledge, this is the first time an integration of the rating credibility, demographic information of users, and ontological semantics of items is created in order to improve the performance of CF recommendation system. Experiments are conducted on the real-world datasets of MovieLens and Yahoo!Movie. Comparing with baseline methods, the experimental results show that the proposed approach significantly improves the quality of recommendation in terms of accuracy, precision, recall, F-measure, and standard deviation of the errors.