Knowledge Enhanced Efficient And Reliable Interactive Recommendation System

Recommendation system is an important means to solve the problem of information overload in social media.In order to solve the problem that traditional recommender systems cannot optimize the user’s long-term experience,researchers have proposed interactive recommender systems.Interactive recommender systems allow users to continuously interact with the recommender system in one session,thereby capturing changes in user preferences in real-time and making better recommendations for users.Researchers have tried to use reinforcement learning to optimize the recommendation strategy of interactive recommender systems.Reinforcement based interactive recommendation systems are faced with efficiency and reliability problems: the efficiency problem refers to the fact that the reinforcement recommendation system faces the problems of sparse feedback,zero learning damages user experience and large item space,etc.;the reliability problem refers to Reinforced recommender systems face problems such as being susceptible to noise interference in practical applications.In view of the above problems,this work introduces knowledge graph into the reinforcement based interactive recommendation system,improves the reinforcement recommendation model through knowledge enhancement and graph contrastive learning,and improves the efficiency and reliability of the interactive recommendation system.The research contents and main contributions of this paper are as follows:(1)In order to solve the efficiency problem of reinforcement based interactive recommendation system,this paper proposes an improved knowledge enhanced policy guided interactive reinforcement recommendation model KGP-DQN.This method builds a behavioral knowledge graph representation module,which combines user historical behavior and knowledge graph to solve the problem of sparse feedback;builds a strategy initialization module,provides an initialization strategy for the reinforcement recommendation system based on user historical behavior,and solves the problem that learning from zero damages the user experience;A candidate set screening module is constructed,and the entire item space is dynamically clustered according to the item representation on the behavioral knowledge graph to generate a smaller candidate set,thereby solving the problem of large action space.The method is tested on three real data sets,and the experimental results show that the method can train the reinforcement recommendation system quickly and effectively and achieve a good recommendation effect.(2)In order to solve the reliability problem of the reinforcement interactive recommendation system,this paper proposes an improved interactive reinforcement graph contrastive learning based recommendation model GCL-DQN based on KGP-DQN.The method constructs a noise data enhancement module which performs data enhancement on the interactive knowledge graph according to the type of noise in the real world;builds a robust node representation module which uses the interactive graph after noise data enhancement to learn the robust node representation of users and items;construct the multi-task learning module which combines the robust node representation module as an auxiliary task with the reinforcement recommendation task as the main task without compromising the performance of the reinforcement recommendation task.This method is tested on two real data sets.The experimental results show that this method can improve the model’s resistance to real noise while ensuring the performance of enhanced recommendation,that is,the model has a certain reliability.