Movie Recommendation Model Fusing Meta-path With Knowledge Collaborative Filtering

The recommendation system has been successfully applied in many fields because it satisfies the needs of individual information screening.The algorithms that are widely used in recommendation systems are mainly collaborative filtering and neural networks.The former has problems such as cold start and interpretability,while the latter has problems such as insufficient recommendation accuracy and interpretability.As a result,scholars began to pay attention to the application of auxiliary information,including knowledge graph.This thesis takes the movie direction of the recommendation system as the research goal,and finds a way to introduce the knowledge graph into the recommendation system.Finally,combined with the idea of hybrid recommendation,a recommendation model fuses meta-path with knowledge collaborative filtering is proposed.The main work of this thesis is as follows:(1)This thesis proposes a meta-path recommendation model.The meta-path recommendation model uses a deep neural network to model the known meta-path reasoning process,and uses a knowledge representation learning method to embed the entities and meta-paths involved.Because the meta-path itself has reasoning interpretability,the meta-path recommendation model is interpretable.(2)This thesis studies collaborative filtering and introduces a knowledge collaborative filtering recommendation model.The knowledge collaborative filtering recommendation model is based on the embedded representation of the knowledge representation learning entity,and compares the similarity between two movies.Because of the recommendation of similar content,it has a higher interpretability in comparison.(3)This thesis proposes a recommendation model that fuses meta-path with knowledge collaborative filtering.It fuses meta-path recommendation model and knowledge collaborative filtering recommendation model.The final model is formed through the linear calculation of the scores of the two models.Because the hybrid recommendation algorithm can integrate the characteristics of each sub-module,the final model is also interpretable.The experimental results show that the recommendation model fusing meta-path with knowledge collaborative filtering proposed in this thesis compares with the recent baselines on multiple evaluation indicators,and has achieved better results.It verifies that the model proposed in this thesis has a certain degree of advancement.