Enhancing signed social recommendation via extracting auxiliary textual information

Real-world applications are increasingly using personalized suggestions to guide users toward interesting content. Graphic Convolutional Neural Network (GCN) has been a great success as a new collaborative filter technology. Nevertheless, the majority of GCN-based systems currently in use can only record historical data about user clicks or transactions, which only reflects a single part of user preferences and item characteristics and ignores numerous important factors like information about the user or item. There is a need to further enrich the potential factors of items from like or related items because goods are not autonomous and may be similar and connected. Moreover, the majority of GCN only function on unsigned networks with only positive linkages available. The process of transferring these models to signed networks is difficult and has received little attention in research despite being extensively observed in practice. In this paper, we suggest a model for extracting auxiliary textual information called Enhancing Signed Social Recommendation Via Extracting Auxiliary Textual Information (E-EATI). By fusing auxiliary information,we can obtain better representation vectors .To combine useful auxiliary information from users or items, We use the BERT model to preprocess the auxiliary textual information and introduce the item2item aggregation operation in our model. Next, we build a signed network using similar and dissimilar individuals of items, By learning the node embedding of the convolutional network of signed graphs, we can get a better embedding representation by integrating different hidden semantic information generated by positive and negative links.In this process, we solve the problem of noise caused by too much information by using similar and dissimilar nodes.We use three standard datasets on on E-EATI : Movielens-1m, Amazon-book, and Yelp2018. The experiment's findings demonstrate that, when compared to the most advanced GCN models, E-EATI significantly improves NDCG and Recall.