Neural Collaborative Filtering Recommendations Enhanced By Social Graphs

With the rapid development of Internet technology,we have ushered in the era of information explosion.While enjoying the convenience of obtaining information,we are also worried about how to quickly and accurately obtain the information we really need from a large amount of information data.In recent years,recommender systems as an effective method to solve information flooding have attracted the attention of a large number of researchers at home and abroad.Among them,the Collaborative Filtering(CF)method has achieved great success by exploiting the user's historical interaction information to mine the user's interest preference.However,CF-based methods usually suffer from data sparsity and cold-start problems.To address these limitations,researchers point out that some side information,such as social network,attributes,context,etc.,can be added to CF.On the other hand,existing research has demonstrated that traditional collaborative filtering methods may not be sufficient to obtain deep semantically rich embeddings.In view of the shortcomings of existing methods,the research contents of this paper include:(1)The model of social recommendation based on multi-task is studied.Aiming at the problems of data sparsity and cold start in CF,we propose a deep learning model that integrates social networks and knowledge graphs.The model mainly includes two modules:recommendation module and knowledge graph embedding module.The recommendation module realizes the user's click-through rate prediction by sharing the user's social feature space.The core goal of knowledge graph embedding is to map entities and relationships in knowledge graphs to a low-dimensional vector space while still retaining their structural information.A multi-task learning paradigm is adopted to exploit the useful information contained in two related tasks together to help improve the generalization performance of the entire model.Through experimental comparative analysis,it is proved that the proposed model has better performance than other baselines.(2)The model of social recommendation based on graph convolution is studied.Aiming at the inability of CF methods to capture the deep semantic information between users and items,this paper proposes a novel end-to-end recommendation framework based on graph convolution.When reconstructing the user-item interaction,the recommendation model explicitly models the high-order connectivity among the user-item,user social network and item collaborative similarity network to improve the embedding representation.By stacking multiple layers of embedding propagation layers,the latent preferences of users in social networks and item collaborative similarity networks are effectively captured.In addition,the feature fusion using the gating mechanism can capture the deep semantic information of nodes.Through experimental comparison and analysis,it is proved that the model proposed in this paper is superior to some existing models in experimental results.(3)An online platform for item recommendation is designed and developed.The platform supports new users to register personal accounts through the registration interface,so as to use the account password to enter and use the platform.Ordinary users can use the social friend association service and item recommendation service provided by the platform.Administrator users can call the background monitoring function to grasp the usage of the system by ordinary users in real time.In summary,this paper exploits social relations from the perspectives of multi-task training and high-order connectivity,and proposes two recommendation models based on deep neural networks.The experimental results show that the proposed model outperforms other models in performance.Finally,based on the proposed model,this paper develops an online item recommendation platform for item recommendation.