Leveraging Socialized Tags To Improve Personalized Recommendation

As an effective tool to solve the problem of information overload,the recommendation system has become an essential part of e-commerce websites and social network platforms.According to the data type,the personalized recommendation algorithm can be divided into two different ways.One is the recommendation algorithm based on explicit scoring data,and the other is the recommendation algorithm based on implicit feedback data.Implicit feedback data is more extensive than explicit feedback data,and the acquisition cost is lower.The use of implicit feedback data for recommendation research can effectively alleviate the user's privacy burden and reflect the user's actual behavioral preferences.However,the interactive data based on implicit feedback only accounts for a very small part of the e-commerce website data and only conveys the user's preference information.The rest is a large amount of missing data.Social tags can reflect user preferences and item features,and can establish a connection between users and their noninteractive items.Therefore,the use of social tag information in implicit feedback data can subdivide the user feedback set and identify the negative samples in the items,which is important to improve the accuracy of the personalized recommendation algorithm.This paper first uses doc2 vec to realize the expression of the user-tag set and the itemtag set in the form of a vector,and through the cosine similarity relationship of the vector,we can help the target user to find the potential preference items.Then,for a user,all items in recommendation system are divided into a positive feedback set,a tag-based potential preference feedback set,and a negative feedback set.After using the tag to subdivide user preferences,this paper constructs two pairs of users' preference relationship hypotheses,and proposes a Bayesian personalized ranking recommendation algorithm based on social tag information called tag-BPR.Experiments on real socialized tag datasets show that the performance of tag-BPR based on socialized tag information has a significant improvement over the classical collaborative filtering algorithm,and the improvement is more obvious in cold start scenarios and data sparse scene.