Social Recommendation System with Multimodal Collaborative Filtering

The widespread use of social network applications has led to an information explosion, making it difficult for users to find target information smoothly and accurately. The personalized recommendation has become a key solution in this situation, and it helps users to easily obtain the information that they are interested in. In this study, we propose a novel framework to provide recommendations more precisely, which is an attention-based recommendation system with multimodal graph collaborative filtering. Our system first exploits the high-order connectivity representations via graph convolutional networks, and then using the relations between users and items to obtain features. After that, our system utilizes attention-based LSTM to capture user's habits and preferred items. By leveraging the user and item information to create multimodal graphs, the proposed system is able to better capture users' habits and provide better recommendations for the users of social network applications. Through experiments on three real-world datasets, we demonstrate that our proposed system is able to significantly outperform the state-of-the-art methods.