Abstract:Aiming at the problems of data sparseness of user item ratings and low recommendation accuracy in traditional collaborative filtering recommendation methods, an improved hybrid recommendation approach based on user interest ratings filling was proposed. Firstly, the users' preference to the item types was analyzed, and the user interest ratings were calculated. Afterwards, the operation of matrix filling was performed. Then the impact of users' subjective ratings differentiation and the item's own quality were considered, and the traditional Pearson correlation coefficient was improved. Based on the filled ratings matrix, users' similarity and items' similarity were computed, to predict ratings from the perspective of users and items respectively. Moreover, the weighted sum of two predicted ratings was calculated further to perform the hybrid recommendation. Finally, experiments were carried out on the Movielens100k dataset. The filling effect of user interest ratings matrix was analyzed firstly, and then the proposed approach, traditional collaborative filtering recommendation methods, and previous methods in the literature were compared and analyzed. The results show that the proposed matrix filling method can effectively alleviate the effect of data sparseness, and the filling effect is better than traditional ratings matrix filling methods. Furthermore, our improved hybrid recommendation approach (IHRIRF) has better recommendations than traditional collaborative filtering recommendation method HCRF as well as WPCC method.