Research On Recommendation Algorithm Based On Knowledge Graph And Graph Convolutional Networ

With the rapid development of social media,personalized recommendation systems model users ’ preferences for items,which can effectively solve the problem of information overload in the Internet era.Traditional recommendation algorithms such as collaborative filtering algorithms face the problems of sparse data and cold start,which limits the development of recommendation systems.A knowledge graph is a knowledge representation model that graphically describes the entities,concepts,attributes,and relationships in the real world.Recently,many scholars have introduced knowledge graph into the recommendation system to model the user ’s interest.By mining the multi-hop relationship(i.e.path)between users and interactive items in the knowledge graph,extracting implicit user preferences and other auxiliary information can help the system learn the similarity between users and items more accurately and improve the recommendation accuracy.However,the existing recommendation systems based on knowledge graph fail to make full use of the global information of the graph to model user interest,and most of them only learn the long-term interest of users,ignoring the influence of time factors on user interest.Most models obtain a fixed-size entity neighborhood structure by calculating the scores between users and relationships,which cannot make full use of the global information in the knowledge graph.The neighbor nodes of the entity are aggregated with the same weight,without considering the different preference of the target entity for different sampling neighbors.Aiming at these two problems,a graph convolution recommendation model combining neighbor node importance sampling and feature cross pooling is proposed.Firstly,the importance of neighbor nodes is obtained by fusing the score of neighbor nodes and the centrality perception score.Then,the feature cross pooling layer is introduced to perform feature cross and aggregation on the target entity vector and the neighborhood vector to obtain the final entity feature representation.Finally,the improved sparrow algorithm is used to optimize the hyperparameters of the graph convolutional neural network.The recommendation performance of the model is verified on three data sets.The results show that compared with the baseline model,the AUC,Recall and F1 indicators of the model in this study have been effectively improved compared with the baseline model.In response to the problem that existing knowledge graph-based recommendation models only model users’ long-term interests and ignore the influence of short-term preferences on users’ future needs,resulting in inaccurate user interests learned in the end,a recommendation model that incorporates similarity negative sampling and users’ short-term preferences is proposed.First,the k-means algorithm is used to cluster the similarity vectors,and the samples with higher similarity to the positive samples are selected as negative samples in each cluster to improve the effect of negative sampling.Then,a graph convolutional network is used to extract features from K sequences of recent user interactions to obtain the corresponding user and item feature vector representations,which are fed into a two-way gated recurrent neural network to learn the short-term interests of users.Finally,an attention mechanism is introduced to assign different weights to different short-term preferences to improve the accuracy of modelling user preferences.The model performance is validated on three datasets,with improvements in AUC,Recall and F1 metrics compared to the baseline model.