Hybrid Algorithm for Item Collaborative Filtering Based on Matrix Factorization

The traditional collaborative filtering algorithm obtains K nearest-neighbor items, which may have the problem of too little similarity among nearest-neighbor items, making the accuracy of rating prediction low. And because in the actual situation, users have very little data on movie ratings, it is less likely for them to rate two movies at the same time, leading to an overestimation of similarity in the comparison. Not only that, users' interest in movies usually changes with time, which affects the prediction of users' ratings of items. To address the above-mentioned problems, this paper proposes a hybrid algorithm for collaborative filtering of movies based on matrix factorization. Firstly, matrix factorization is used to fill the vacant part of the user-item rating matrix, which is incorporated into the K nearest neighbors of the collaborative filtering algorithm, effectively alleviating the problem of low similarity between nearest neighbor items and improving the accuracy of predicted ratings. Secondly, a common rating weight for rating differentiation is added to calculate the similarity between movies to correct the bias of the similarity calculation. Finally, a temporal weighting function is introduced in predicting ratings, thus reducing the influence of temporal factors on user preferences and further improving the accuracy of recommendations. Experiments on the Movielens dataset show that the algorithm has significantly improved over traditional collaborative filtering algorithms.