Research On Knowledge Graph Recommendation Algorithm Based On Associated Path Reasoning

With the rapid development of the Internet,people often face the problem of information overload when choosing information.In order to help people to dig out the interested information in the large and scattered information,recommender system has become a research hotspot in academia and industry.Knowledge graph contains rich entity structure and semantic association information.Knowledge graph can be used as auxiliary information for recommendation algorithm to reduce data sparsity and cold start problem.In the traditional recommendation algorithm research,collaborative filtering algorithm relies on the interaction matrix between user and item to recommend the item that the user is interested in.Actually,compared with a large number of users and items,the interaction between users and items appears to be very sparse.The data sparsity problem makes the recommendation effect of traditional collaborative filtering model not ideal.Moreover,traditional recommendation systems do not provide reasons for users to recommend,and the opacity of recommendation models reduces users’ trust and satisfaction with them.Considering the above problems,this paper proposes a knowledge graph recommendation algorithm based on associated path reasoning,takes account of user’s interest decay factor,and supplements the rich entity structure relationship and semantic information in the knowledge graph,so as to realize personalized recommendation and give recommendation interpretation.The model proposed in this paper integrates rich node semantics and inter-node relationship information in the knowledge graph,which can effectively alleviate the data sparse problem existing in the traditional collaborative filtering algorithm and improve the recommendation performance.The model framework proposed in this paper includes the associated path search module and the recommendation module,and the main innovative contents are as follows:(1)In the associative path search module,the breadth first soft matching algorithm adopts the spontaneous search mode to avoid the problem that the traditional interpretable recommendation model induces path rules to rely on human experience.(2)The associated path search module gives the recommendation model the possibility of explanatory capabilities.Based on the breadth first soft matching algorithm,the associated path search module searches the effective associated path between user and item according to the node access probability,fully considers the structural location information of nodes in the knowledge graph.The associated path is used as the basis for the recommendation to reasonably infer the reason why the user chooses the item.(3)The recommendation module is based on the associated path search module.When the user’s interest in items propagates along the associated path,it will fully consider the degree of interest decay at different distances between different items,and learn the associated path feature as the user feature according to the interest decay factor,and calculate the user interest preferences of different items to discover the potential interests of users.In order to verify the performance of the model,this paper constructs the corresponding knowledge graph on two real data sets of Amazon,and conducts model experiments.Through sensitivity analysis of key parameters and comparison with other baseline models,the robustness and effectiveness of the proposed model are verified.Simultaneously,the reasoning process of recommending items for users is intuitively demonstrated through a recommendation case,and the final recommendation suggestion is given to verify the interpretability of the model.