Graph Convolutional Multi-task Recommendation Model Based on Feature Enhancement
Li Guoxuan
Luo Weiqun
Luo Lijin
Zhang Zijian
Lu Jingwei
Abstract：The advanced capabilities of Graph Convolutional Neural Networks(GCNs)in message aggregation and propagation have significantly improved the accuracy of knowledge-graph-based recommendation systems.Despite this,many current knowledge graph convolution methods haven't deeply explored users'interests and fail to effec-tively model triplet relationships for the convolved entity features.In response to these limitations,this paper propos-es a multi-task graph convolutional recommendation model based on feature enhancement.The model first uses the BERT pre-trained model to model the initial representation of entities,and then propagates and updates node features through GCNs embedded with an attention mechanism.Furthermore,a triplet relationship modeling strategy is intro-duced to perform representation learning on the updated entity vectors.In the prediction stage,a Deep Neural Net-work(DNN)is used to fuse the feature vectors of users and entities to achieve more accurate recommendations.To verify the effectiveness of the model,extensive experiments have been conducted on common public datasets in the recommendation field.The experimental results show that compared to the KGCN model,our model achieved a 1.5％and 1.3％increase in AUC and F1 on the Movielens-20M dataset,a 1.8％and 1.0％,increase in those on the Book-Crossing dataset,and a 4.8％and 3.6％increase in those on our self-built dataset,respectively.These significant per-formance improvements demonstrate the superior performance and application potential of our model in the recom-mendation system field.
Keywords：Knowledge GraphMulti-taskRecommendation AlgorithmPre-trained ModelGraph Convolutional Neural Network(GCN)
Publication Date：2025-02-28
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：9( 72-80 )