Abstract: In the recommendation model based on graph convolution network,the graph convolution only aggregates information from the input nodes with identifier information, which will decrease the recommendation precision and, thus, lead to a bottleneck problem. To solve this problem,a lightweight graph convolution network recommendation model based on text information fusion is proposed by considering enriching node features with auxiliary information. The model extracts text comment features from convolution neural network and adds them to the node embedding of graph. To simplify the structure of graph convolution network,the proposed lightweight graph convolution framework is used to transmit the fused feature information linearly on the user-movie item graph to learn the embedding of the user and movie item. The weighted sum of all sub-levels of the graph convolution is used as the final feature output for predicting the rating. Experimental results on three real datasets show that the proposed method can alleviate the bottleneck problem of information aggregation and improve the accuracy of recommendation. The model can also alleviate the cold start problem.

Key words: recommender model, information aggregation, graph convolutional network, text information