Research And Application Of Personalized Recommendation Algorithm Based On Knowledge Grap

In the era of information overload,recommendation algorithms help users find the information they need accurately and quickly,but most of today’s recommendation algorithms were based on user behaviour data and item content information,which can produce useful recommendations but still suffer from problems such as sparse data,poor interpretability and cold starts.How recommendation algorithms can solve these problems was the current focus of researchers,and the main research of this topic was as follows:(1)In order to organise the valuable information accumulated in the domain into knowledge that can be reused,this topic proposes a method for to construct the knowledge graph of the online education vertical,thus build the knowledge graph of the online education vertical.An ontology base was constructed by analysis of domain knowledge,extracted knowledge related to online education from the dataset,instantiated the ontology base,and stored the constructed knowledge graph in the graph database.(2)To address the problem that the representation learning model for knowledge graph can hardly balance efficiency and results in large knowledge graph,this topic proposes a representation learning model CTrans D for knowledge graph,which used cluster algorithm to cluster entities to reduce the number of entity projections,reduce the spatial complexity of the model,convert the similarity between entity classes into a probabilistic representation so that similar entities have similar projection vectors after cluster.The model was trained to be a better co-optimisation model,and finally experiments were designed to verify the performance of the model.(3)In order to address the problems of sparse data,cold start and difficulty in capture the preference of different users and items relationships in existing recommendation algorithms,this project proposes a recommendation model CTrans DGAT,which combines knowledge graph and graph attention network.The knowledge graph was accurately mapped in a low-dimensional vector space by using a representation learning model,based on which a joint graph attention network was used to propagate and aggregate information from nodes based on weight coefficients,obtain recommendation results by score prediction,and finally design experiments to test the performance of the model.(4)In order to make the recommendation model have better practical application,this project designs and implements a personalised recommendation system for online education,in which the proposed recommendation model was applied to make personalised recommendations for courses.