Resource Recommendation System Based On Domain Knowledge Graph

With the continuous development of informatization,people come to realize the knowledge shared among the network is on an unprecedented scale.Thus it brings about a drastic increase in the amount of data.Recommendation system has been produced and gradually developed,and has achieved good results in some application fields.However,the continuous digitization of educational resources has led to a very messy type of resources returned by users' searching for knowledge from the internet,and most of the existing recommendation systems in the education field recommend single-type resources.Based on the above problems,this paper designed a recommendation system to complete the collection and recommendation of resources in the education field,helping users to inquire about domain concepts and recommend related resources.Combining the current recommendation method and education direction,this topic takes the computer field as an example,through the integration and classification a variety of computer field resources,linking the resources with the concepts in CSO(computer domain ontology),designing two recommendation methods based on the knowledge graph and completing the implementation of the system.The main research contents are as follows:1)A knowledge graph of computer resources is created.Based on the existing domain ontology,first,the crawler technology is adopted to obtain domain-related resource data to data preprocessing.After that,the resources are classified according to the domain concepts in the ontology.Finally,the resource data corresponding to the concept is linked to the ontology as attributes,which provides semantic information about recommendation.2)A weighted recommendation method based on the neighborhood expansion features of the knowledge graph(KGTR)is designed.For users with existing interactions,according to existing algorithms and the recommendation model fused with knowledge graph,the relationship strength of the entity is added to the knowledge graph convolution network(KGCN),and it is combined with the results of the extended users' interest features model(Ripple Net)that expands users' features using a weighting method,at the same time the KL divergence(Kullback-Leibler)is introduced as the deviation to improve accuracy of the recommendation results.3)A user's cold-start recommendation method combining users' attributes and implicit feedback is designed(User CF-SIM-Node2vec).For users without interaction history,based on the method of User CF,the similarities among users are calculated through implicit users' behavior and contributes to obtaining neighbor users.At the same time,the graph embedding method is used to obtain the characteristics of the domain resources,which is used to obtain more accurate similar resources,thereby alleviating the user's cold-start problem.4)A computer resource recommendation system based on the domain knowledge graph is designed.Users can query about computer-related concepts and resources through the system,and obtain personalized resource recommendations.