Research On Personalized Information Recommendation Algorithm Based On Social Data

With the rapid development of Internet technology and social platforms,the amount of information that users receive on a daily basis has increased dramatically.Personalized recommendation algorithms can effectively alleviate the problem of information overload.Among them,social recommendation can recommend the most interesting information to users through data of group tags and friend interactions.However,in the existing social recommendation methods,problems such as data sparsity and low recommendation accuracy do exist.Previous studies have paid more attention to explicit feedback data such as review ratings,and often ignored the influence of implicit feedback data such as social trust relationship on user interest preferences.In response to these problems,this thesis first proposes a multi-layer attention recommendation model based on the social trust relationship(Social Relationship Recommendation Model with Multilevel attention,MA_SRec),to solve the problem that deep and shallow social trust relationships have variant degrees of impact on users’ interest preferences.The model uses the LSTM long and short-term neural network to extract the user’s dynamic trust network,and learns the influence weight of the recent social trust relationship by the attention mechanism.Assigning respective influence to features can accurately describe the user’s interest preferences.Concerning the fusion of users’ original features and social relationship features inspired by the transfer learning method,this thesis innovatively proposes a knowledge transfer recommendation model(TR_SRec)based on social trust relationship.This model takes the social feature domain as the source domain,and user features as the target domain,and map the richer auxiliary information from the social relationship domain to the user feature domain by means of knowledge transfer,so as to more accurately describe the user’s interest preferences,effectively alleviate data sparseness,and increases the interpretability of the model.This thesis draws on the real information and social data set Epinions,and compares MA_SRec and TR_SRec with classic algorithms such as social algorithm SErec,SBPR,and implicit feedback recommendation algorithm WRMF.The experimental results show that MA_SRec is basically the same in MRR@10,and the other three indicators have improved to some extent.The four indicators of TR_SRec have improved significantly.The experimental results suggest that the two methods both conduct in-depth analysis of social information and effectively decrease the problem of data sparsity.The model supports application in information recommendation and various social recommendation scenarios.