Research And Application Of Personalized Recommendation Algorithm Based On Knowledge Graph

With the development of information technology and the popularization of mobile internet,the amount of data people are facing is rapidly increasing.Personalized recommendation algorithms can filter out potentially interesting information from a massive amount of information,effectively alleviating the problem of information overload.The recommendation algorithm based on the knowledge graph integrates the knowledge graph as auxiliary information into the recommendation algorithm,which can improve the accuracy of the recommendation algorithm and the diversity of the recommendation results,overcome the data sparsity and cold start problems existing in the traditional collaborative filtering recommendation algorithm,and has been widely studied in recent years.However,existing knowledge graph based recommendation algorithms suffer from insufficient utilization of collaborative and knowledge graph information,and generally adopt supervised learning paradigms,resulting in sparse supervised signals.In addition,the noise problem in implicit feedback data further affects the accuracy of recommendation algorithms.This thesis conducts research on the above issues,and the main contributions of this thesis are as follows:1.In order to solve the problem that the existing recommendation algorithms based on knowledge graph do not make full use of collaborative information and knowledge graph information,we proposed a recommendation algorithm based on graph neural network and knowledge graph propagation.The light graph convolution neural network is used on the user-item interaction graph to obtain the collaborative information of users and items,and the propagation based method is used on the knowledge graph to obtain the semantic information of users and items.The experimental results on three public datasets show that the performance of the recommendation algorithm has been effectively improved.2.We proposed a recommendation algorithm based on graph neural network and knowledge graph propagation to address the sparse supervised signals and noisy useritem interaction data in knowledge-aware recommendation algorithms.Contrastive learning is introduced as an auxiliary task into the knowledge-aware recommendation algorithm,and the robustness of the model to noisy data in interactive data is improved through data augmentation and contrastive learning,by using collaborative contrastive learning to improve the insufficient model learning caused by sparse supervised signals,experimental results on three public datasets show that contrastive learning can effectively improve the performance of recommendation algorithms.3.We designed and implemented a knowledge graph based book recommendation system,and applied the knowledge graph based recommendation algorithm proposed in this thesis to practical applications.The main functions of the system include user registration and login,personalized book recommendation,and popular book recommendation.