Recommendation Models Fusing Multi-source Heterogeneous Data Based On Social Relationship

The rapid development of the information age has accumulated a large amount of data in the network,and information overload has become an urgent problem to be solved.The recommendation system came into being in order to solve this problem.Due to a single recommendation model cannot handle the heterogeneous data,researchers try to fuse different models and propose a hybrid model to provide recommendations for users.However,most hybrid recommendation models only fuse data from the perspective of recommendation results,the degree of data fusion is low.Depending on the choice and design strategies by researchers,there are problems such as low recommendation accuracy,cold start and poor scalability in most hybrid recommendation models.Social relationships play an important role in people's daily lives.Users are usually affected by their direct friends'or even indirect friends'preferences.Users are more likely to choose items that their friends have purchased.The recommendation of the user's friend is beneficial to increase the user's trust in the recommendation result.Based on this,this paper proposes a hybrid recommendation model(BRS_cS,BPR-Review-Score-Social)based on social relationship that fuses multi-source heterogeneous data,constructs a unified representation learning framework,and integrates heterogeneous data such as ratings,comments,and social relationships from the data source level.The BRS_cS model introduces social relationships into the recommendation system through the user friend trust model,uses the Paragraph Vector-Distributed Memory model to process review data,uses the fully connected neural network to process rating data,and uses an improved Bayesian Personalized Ranking model to optimize the ranking results.In order to enhance the scalability of the recommendation system,this paper also proposes a scalable hybrid recommendation model(s BRS_cS,scalable BRS_cS).When introducing new data into the s BRS_cS model,there is no need to redesign the recommendation framework and retrain the model parameters.In this paper,the models are verified on the Yelp public dataset.The experimental results show that the BRS_cS model and the s BRS_cS model are superior to other recommendation models in terms of accuracy,recall,and normalized cumulative loss gain.The introduction of social relations improves the recommendation quality and solves the cold start problem,makes the recommendation system more interpretable.The proposed s BRS_cS model enhances the scalability of the recommendation framework.In addition,this paper also obtains the best parameter setting through experiments.Therefore,models proposed in this paper can improve the accuracy of the recommendation results,solve the problems of cold start and data sparseness to a certain extent,and make the recommendation system more scalable.