| People’s lives are made more convenient by the Internet’s quick expansion,yet the amount of data available is also expanding at an exponential rate.Users may get rich data information,yet the abundance of data information may overwhelm users.The issue of data information overload is efficiently resolved by the introduction of a recommendation system.But still,the knowledge graph contains rich and fine-grained information,in which semantic association helps to mine potential connections,due to the issues of sparse data and cold start in traditional recommendation algorithms.The ability of interpretable recommendations may be effectively improved by employing the knowledge graph as supplementary information in the field of a recommendation system,which can then offer a more precise personalized recommendation.However,when the entity is integrated from the knowledge graph to represent semantic information,the existing knowledge graph-based recommendation models do not fully take into account the semantic information of users and items,making it impossible to accurately reflect the connectivity between users and items.As a result,the process of providing interpretable recommendations by relying on the topology of knowledge graphs is not fully reflected.In light of the issues at hand,this study fully exploits the structural properties of knowledge graphs to extract potential user interests and preferences,explore and expand the potential semantic information of items,improve the semantic representation of items,establish a stronger connection between users and items,and provides interpretable personalized recommendations.The following are the primary research contents covered in this thesis:(1)This thesis proposed a recommendation model KGERS based on knowledge graph to enhance the semantic representation of items.The higher-order collaboration information of items from the user-oriented entity view and the higher-order knowledge information of items from the item-oriented entity view are learned utilizing heterogeneous link propagation strategy since the existing recommendation model cannot fully utilize the topological structure of the knowledge graph to enhance the semantic information representation of items.The collaboration information and knowledge information are combined using the knowledge-aware enhanced attention representation mechanism to acquire the potential semantic information of the item and enrich the semantic information representation of the item.Three open-source datasets are used for experimentation and verification.The KGERS model’s AUC and F1 in the CTR prediction scenario and Recall@K in the Top-K recommendation are both better than the baseline models currently being used.(2)Aiming at the problem that KGERS model cannot provide high order interpretable recommendation and personalized recommendation,a scoring weighting method based on connection relation and user information representation was proposed,and the scoring function of personalized relation was introduced,which could screen candidate items that fit attribute characteristics according to user preferences and provide personalized recommendation for users.More importantly,the relationship types that users are not interested in will be given less weight,which filters out some interference entities and triples to a certain extent,and theoretically constructs a more accurate interpretable path to achieve interpretable recommendations.And the method is integrated into the KGERS model to make it have the ability of interpretable recommendation and personalized recommendations.(3)Based on the research work,a prototype system with movies as recommendation content is designed and implemented.The system’s structure and functional modules are designed based on an analysis of the system’s requirements in order to properly propose high-caliber movie content to users.Combined with actual application scenarios,the practical application value of the research contained in this thesis is verified. |
PDF Full Text Request