Research On Recommendation Algorithm Based On Knowledge Graph

Recommender system,as an effective means to filter out the contents that users are interested in from massive information,has rich research and practical value.In recent years,using auxiliary information to help recommendation has been widely studied,among which the recommendation algorithm based on knowledge graph is one of the focused areas.However,through the research of this thesis,it is found that there are still some issues can be improved for the recommendation algorithm based on knowledge graph.First of all,the existing methods do not consider the information of item granularity when propagating the user's historical records on the knowledge graph,which leads to the lack of information and does not distinguish the importance of different historical items of the user;secondly,existing methods do not take into account the correlation information between entities within the user entity sets and item entity sets,as well as the interaction between the user entity sets and the item entity sets.This thesis studies the above issues and proposes two recommendation algorithms based on knowledge graphs to solve these issues.The main contributions of this thesis are as follows:1.Conduct research on the existing knowledge graph based recommendation algorithms and related technologies such as deep learning and attention mechanism.Analyze and summarize the research status and shortcomings of knowledge graph based recommendation algorithms.2.Propose Interactive Knowledge-aware Attention Network for recommendation(IKAN).Existing knowledge graph based recommendation methods do not consider the information of item granularity when propagating the user's preferences on the knowledge graph,and do not distinguish the importance of different historical items of the user.When the algorithm proposed in this thesis is used to propagating the user's historical records on the knowledge graph,each item is modeled separately,which can catch information of item granularity.The representation of the item in the user's history record and the target item enhanced by knowledge graph can be obtained.Through interaction aware attention network to distinguish the importance of different historical items of the user,better user representations can be obtained,thereby improving the accuracy of recommendation algorithms.3.Propose Knowledge-aware Entity Inner-Outer Interaction Network for recommendation(KEIOIN).Aiming at the problem that the existing methods do not consider the inner interaction between the entities within user entity sets and the item entity sets and the interaction between the user entity sets and the item entity sets,the algorithm in this thesis respectively propagates the user and the item on the knowledge graph.Based on the idea of self-attention mechanism,the interaction between entities within the entity set can be catched by relation self-attention network,and then combine the collaborative attention network that models the interaction between the user entity sets and the item entity sets to form an inner-outer interaction network structure.At the same time,the interaction within the entity sets and the interaction between user entity sets and item entity sets are both learned,thereby improving the accuracy of the recommendation algorithm.4.The two knowledge graph based recommendation algorithms proposed in this thesis are tested on multiple real world data sets.Through the summary and analysis of the experimental results,the effectiveness of the proposed algorithms can be verified.