Research And Implementation Of Entity Recommendation Model In Heterogeneous Information Network

The recommendation system can help people quickly find the information they need from the Internet.The traditional recommendation system only takes into account the single type of relationship between the user and the item.According to the different types of input data,the recommender system is divided into two kinds:the input data in the recommender system is users' explicit feedback(e.g.rating,etc.)and the input data of the recommender system is users' implicit feedback(e.g.viewing behavior,purchasing behavior,etc.)In the Internet industry,there are abundant types of users' implicit feedback,and using the implicit feedback can achieve better recommendation than only using users' explicit feedback.In order to improve the recommendation effect,this thesis proposes two entity recommendation models in heterogeneous information network(HIN),which combines multiple types of links between users and items and various types of implicit feedback of users.One of the two is to learn the importance of implicit features on different meta paths by combining the different semantic meta paths with users' multiple types of implicit feedback.According to this,we get the users' preferences for items,and then we can use the users' preference to make recommendation.The model is called a entity recommendation model by integrating multiple types of implicit feedback based on heterogeneous information network(Mult-heterec).Another entity recommendation model is a personalized recommendation model(Mult-heterec-p)that combines users' multiple types of implicit feedback and groups' interests.Base on the basis of Mult-heterec model,we first cluster users to find groups with similar behaviors,and then learn Mult-heterec model for users in each group.The two entity recommendation models based on heterogeneous information network can achieve better recommendation in quality,and the two models both can effectively solve the problem of data sparsity and cold start in collaborative filtering recommendation system.The innovation points of this thesis are as follows:1.This thesis proposes a entity recommendation model(Mult-heterec)by using users multiple types of implicit feedback in heterogeneous information network,which improves the performance of the recommendation and can effectively solve the problem of data sparsity and so on.The Mult-heterec model is constructed by learning the importance of users' preference implicit feature under different meta paths.And we use Bayesian Ranking Optimization algorithm to learn the parameters in the model.The Mult-heterec model makes recommendation according to the value of user-item preference score matrix.2.This thesis presents a personalized recommendation model by combing users' multiple types of implicit feedback with the interest of groups.First,we use Kmeans clustering algorithm to cluster users'multiple types of implicit feedback to find groups in which users have similar 'behaviors.That is,the users who have similar behavior have similar interests.Then we learn the Mult-heterec model for the users in each group,and make recommendation to the users.3.In this thesis,we use the Douban dataset which is a real dataset to verify the effects of the proposed model.Form the experimental results,we can see that both Mult-heterec model and Mult-heterec-p model can improve recommendation performance,and the recommendation effects of Mult-heterec-p model is better than the Mult-heterec model.