Matrix factorization recommendation algorithms fusing weighted heterogeneous information network
WANG Gensheng,Pan Fangzheng
(School of Humanities, JiangXi University of Finance and Economics, Nanchang 330013, China)
(School of International Trade and Economics, JiangXi University of Finance and Economics, Nanchang 330013, China)
[Objective] In order to improve the quality of recommendation, the knowledge of weighted heterogeneous information network is integrated into the matrix decomposition recommendation algorithm.
[Methods] Firstly, the heterogeneous information network is constructed, and the weight of connection is calculated by the improved tanh function; Secondly, the meta paths are selected in the network and their weights are calculated by information gain; Then, the user interest similarity under different meta paths is calculated, and then the user similarity matrix is calculated by combining the weights; Finally, the user similarity matrix is integrated into the objective function of the matrix factorization recommendation algorithm.
[Results] The experimental results on the Hetrec2011-MovieLens-2k dataset show that, compared with the traditional matrix factorization recommendation algorithm (FunkSVD), the accuracy, recall and coverage of the improved algorithm are increased by 4.4%, 5.4%, 4.6% respectively, and the root mean square error (RMSE) is reduced by 0.6.
[Limitations] When faced with massive data, the efficiency of matrix decomposition is low, and the problem of user interest drift is not considered.
[Conclusions] The algorithm proposed in this paper improves the recommendation quality of matrix decomposition recommendation algorithm.