| The integration of information technology and education has led to online learning becoming a common way of learning.Compared with the traditional offline learning,online learning has no limitation of physical classroom,and can break through the constraints of time and space.While online learning has brought great convenience to learners,it also causes problems such as overload of learning resources and difficulty in selection.Recommendation system is an important method to help learners accurately select relevant and useful information,but the recommendation method based on online education is still facing severe challenges:first,the existing online educations platforms have the problem of data sparsity.Second,the mining of implicit interaction data in traditional recommendation algorithms is not sufficient,and the guiding role of knowledge behind recommendation is not obvious.Therefore,how to mine learners’ preferences has become a current research hotspot.Knowledge graph can be used to imply the potential relationship between items.Mapping learning resources and their attributes into the knowledge graph can better understand the potential connections between them.It also makes the recommendation results interpretable.In order to solve the above problems and improve the performance of learner personalized recommendation,the main work of this paper is as follows:(1)Aiming at the cold start problem caused by data sparsity in personalized recommendation of learning resources,this paper proposed a Learner Personalized Resource Recommendation Method Based on Knowledge Graph(LPRM).The LPRM model uses the historical interaction information between students and courses in online learning and the attribute information of online courses to construct the course knowledge graph to assist the personalized recommendation of course resources.Aiming at the problem that the propagation of entity relationships in Ripple Net framework does not consider the influence of entities,a node influence calculation model was proposed to measure the influence of entities in the knowledge graph.The LPRM model framework was constructed to obtain learners’ ratings of learning resources.A large number of comparative experimental results verify the superiority of this model,and the results of model parameter analysis show that the LPRM model can effectively improve the performance of learners’ personalized learning resource recommendation.It also better alleviates the inaccuracy of learners ’personalized learning resource recommendation caused by data sparsity.(2)Aiming at the problem that the random selection of neighborhood in the existing knowledge graph convolutional network recommendation model limits the effect and stability of the auxiliary recommendation of knowledge graph,this paper proposed a knowledge graph convolutional network recommendation model based on the importance ranking sampling method of structural holes and public neighbors sorting(KGCN-SHPN).Firstly,SHPN sampling method is used to sort and sample the receptive field of each entity in the knowledge graph.Then,the entity information and the information collected from the neighborhood of the entity are aggregated according to the graph neural network to obtain the feature representation of the learning resource.Finally,the feature representation of the learner and the feature representation of the learning resource are obtained according to the prediction function to obtain the interaction probability.Experiments are carried out on three data sets.The experimental results show that the proposed model is superior to KGCN,Ripple Net and other knowledge graph-based recommendation models in terms of evaluation indicators AUC and ACC,which proves the superiority of the model proposed in this paper. |
