College of Mechanical and Electronical Engineering, Northeast Forestry University, Harbin City, Heilongjiang, China

---

Abstract—This paper proposes an effective recommendation algorithm based on multi-source information, which employs the user feature information and image feature information to handle the problems in recommender system, such as data sparsity, cold start user problem and cold start item problem. The proposed algorithm is as follow. Firstly, this paper presents a denoising auto-encoder to handle the problem of data sparsity and cold start user problem. It can learn the hidden features with nonlinearity of user and item. In addition, the paper proposes collaborative filtering algorithm based on multi-features of items. This approach employs the convolutional neural network to extract features of the image. Then combine the features of the image and the activities of the users to solve the problems of data sparsity and cold start item problem. The proposed method mentioned above is tested with dataset called MovieLens. The results of the experiment show that the proposed method has competitive performance

Index Terms—Recommendation algorithm, denoising auto-encoder, convolutional neural network, data sparsity, cold start

[PDF]

Cite: Lina Tang, Keqi Wang, "An Effective Recommendation Algorithm Based on Multi-Source Information," Journal of Software vol. 14, no. 3, pp. 107-115, 2019.