Research And Implementation Of Personalized Recommendation Algorithm Based On Multi-aspect Social Influence

In recent years,social relationship,as a novel data type,has injected new vitality into the development of recommendation system,thus the research direction of social recommendation is derived.Social recommendation models aim to maximize the auxiliary role of social information and build recommendation systems with higher accuracy.Most of them are based on homogeneity and social influence theory,that is,users’ preferences are assumed to be similar to or be influenced by their friends.However,the social influence from friends can be strong or weak inherently,and the strength is multi-aspect and dynamic according to different recommendation contexts.Most of the existing studies only considered the strength of social influence without in-depth consideration of its other characteristics.Therefore,there is still a broad space to explore how to integrate various features of social influence into the recommendation model and provide users with more personalized recommendation experience.Based on this background,this paper mainly studies the personalized recommendation algorithm based on multi-aspect social influence,and focuses on exploring the multi-aspect and dynamic nature of social influence.The main work is summarized as follows:(1)one recommendation model based on multi-aspect social influence(CapsRecI)is proposed and implemented.Based on capsule network,this model adaptively extracts multiple social influence vectors from users’ social relationships by using dynamic routing method.When facing a specific recommendation context,the social influence from the optimal aspect is dynamically matched according to item characteristics.(2)Based on previous model,a new improved model(CapsRecII)is proposed and implemented by integrating the user behavior information modeling.In this model,users and items are modeled separately by graph attention network,and the original graph node feature aggregation method is improved based on the consideration of bias effect and neighbor node interaction.(3)A recommendation system is designed and implemented in the field of open-source code repository,and the system uses the model proposed in this paper to recommend open-source repositories for users.In order to verify the validity of the proposed model,we conducted experiments on public datasets.The experimental results show that the model proposed in this paper performs well and has obvious improvements compared with other comparison models.In addition,after practical tests,the code warehouse recommendation system designed and implemented in this paper works well,which further verifies the application value of the model proposed in this paper in practical scenarios.