Graph Convolution Recommendation Algorithm Integrating Multi-relationship Preferences

With the continuous development of Internet technology, existing data forms are becoming more and more complex. However, traditional recommendation algorithms have limitations when processing complex structured data. This paper improves the message propagation layer by considering user social relationships and item similarity relationships. Firstly, An item similarity calculation method that combines user activity is proposed, which combines social relationships and improved item similarity information with interaction information to represent the embedding of users and items. Secondly, the attention mechanism is integrated in the process of aggregating information to make the model more interpretable. Finally, high-order transfer of information is achieved by stacking graph convolutional layers. And perform inner product operation on the user and item representations obtained in the propagation layer to achieve prediction. Experimental verification was conducted on the Lastfm data set. The experimental results show that the performance of this method is the best, proving the effectiveness of the method.