Semantic Preference-based Personalized Recommendation On Heterogeneous Information Network

The Internet has become an indispensable part of people's lives in recent years,however,it is difficult for Internet users to quickly find out the information they were interested in facing the massive data on the Internet with the explosive growth of data volume in the era of Internet information today.In this context,in order to help users find the information of interest more conveniently and accurately,the recommendation system algorithm emerged.It records and analyzes the user characteristics based on the user's historical records to obtain the corresponding user portrait.Then it will meet the user's preferences to the user using the attributes of the information or the dependency between them to recommend information.Traditional collaborative filtering methods have been widely used in many recommended technologies.These methods use the historical information of the interaction between the user and the commodity to construct the user's feature model,and adopt the users with the similar feature model to conduct mutual recommendation.However,in practical applications,users only have a limited number of interaction records and products frequently,which lead to the data sparsity problem of the algorithm.Now,sophisticated data mining technology was able to analyze the increasingly comprehensive information on people's personal preferences as well as commodity attribute information.Consequently,many researchers have used external information to improve recommendation technology.However,most previous studies consider only adding single relationship types,such as social networking friend-relationships.In the real world,considering multiple types of external relations can more accurately determine the reason why a user selected an item.To address this problem,in this paper,we propose a hybrid method called the semantic preference-based personalized recommendation on heterogeneous information networks(SPR),which combines user feedback scores with heterogeneous information networks.This method can improve recommendation problems by considering multiple types of external relationships.To apply the method,we first introduce a similarity measure between users based on a user's potential preferences in the meta-path and design the recommended model at the global and individual level.Finally,we perform experiments on two real-world datasets,finding that the SPR method achieves better results compared to several widely employed and state-of-the-art recommendation methods.Our contributions in this research are summarized as follows:1.We use rating data which users rated items to carry out binary transformation in heterogeneous information network,at the same time,we utilize the scores to filter the high score items,and construct the heterogeneous information network based on the binary data.2.We utilize the meta-path in heterogeneous information networks to represent the preferences of the user and construct the user's preference model.And we calculate the similar measure between users based on the semantic interpretation of the meta-paths.Lastly,we use the relationship between similar users to construct the target user's personalized recommendation model.3.Our studies utilize two real-world datasets,Douban and Yelp,to demonstrate the performance of our method.