A Group Recommendation Approach Based on Neural Network Collaborative Filtering

At present, the most popular recommendation algorithms belong to the class of latent factor models(LFM). Compared with traditional user-based and item-based collaborative filtering methods, the latent factor model effectively improves recommendation accuracy. In recent years, deep neural networks have succeeded in many research fields, such as computer vision, speech recognition, and natural language processing. However, there are few studies combining recommendation systems and deep neural networks, especially for group recommendation. Some academic studies have adopted deep learning methods, but they mainly use it to process auxiliary information, such as acoustic features of sounds, and semantic analysis of texts, the inner product is still used to deal with latent features of users and items. In this paper, we first obtain the nonlinear interaction of latent feature vectors between users and projects through multi-layer perceptron(MLP), and use the combination of LFM and MLP to achieve collaborative filtering recommendation between users and items. Secondly, based on the individual's recommendation score, a fusion strategy based on Nash equilibrium is designed to ensure the average satisfaction of the group users. Our experiments are conducted on the Track 1 of KDD CUP 2012 public data set, taking the square root mean square error(RMSE) as the evaluation index. The experiment compares the traditional LFM optimization model, the MLP model and the LFM-MLP hybrid model in individual recommendation, and compares the strategy proposed in this paper with the traditional three single group strategies, the most pleasure, the average strategy and the least misery. The experimental results show that the proposed method can effectively improve the accuracy of group recommendation.