A Group Recommendation Approach Based On Neural Network Collaborative Filtering

Nowadays,people perform frequent social activities on social networks such as Weibo,and users can be divided into groups according to user preferences.The recommendation system can effectively help users filter irrelevant information,find information of interest to them,and improve the utilization of information by analyzing and mining relevant information about collecting users and commodities.At present,the most popular recommendation algorithm belongs to the latent factor model(LFM).The LFM decomposes the original rating matrix into the product of two low rank feature matrices and uses the inner product to obtain the predicted score.Compared with traditional user-based and item-based collaborative filtering methods,LFM 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.This thesis uses micro-blog data to conduct research,explores collaborative filtering combined with neural networks,and completes group-based friend recommendation,which is very meaningful in the field of social recommendation.This thesis is based on the linear interaction of the latent factor model to obtain the linear interaction relationship between the latent feature vector between users and items through the inner product.And then the multi-layer perceptron(MLP)is used to obtain the nonlinear interaction of the latent feature vector between users and items.A combination of a factor model and a multi-layer perceptron to achieve collaborative filtering recommendations between users and items.Secondly,in order to ensure the maximum satisfaction of group members,this thesis proposes a fusion strategy based on Nash equilibrium after obtaining individual recommendation ratings.The strategy is applied to Weibo user group recommendation to reduce the root mean square error(RMSE)of the group recommendation.The main work of this thesis is as follows:(1)The individual recommendation hybrid model proposed in this thesis uses the neural network method based on the linear interaction information between users and items in the LFM.This thesis constructs the MLP model of users and items in the latent factor space,which is used to obtain the nonlinear interaction between the latent feature vectors of users and commodities.Combining the two sub-models,the recommen-dation accuracy of the LFM is higher than that of the traditional LFM.(2)In the group recommendation for Weibo users,this thesis studies and analyzes the traditional fusion strategy,draws on the idea of game theory,and transforms the problem of solving the best group satisfaction into the problem of finding Nash equilibrium.Compared with the traditional fusion strategy,the proposed Nash equilibrium fusion strategy reduces mean value of RMSE in group recommendation,and the recommendation effect is better.(3)The experimental results of the individual recommendation show that the proposed model improves the recommendation accuracy by 3.17% compared with the improved LFM model.The experimental results of the group recommendation show that the proposed Nash equilibrium fusion strategy is 6.83% higher than the traditional mean strategy.