Potential purchase intention capturing method based on consumer's interests mining in social networks

Traditional session-based recommendation systems usually capture users' potential purchase intention from their recent transaction records, which does not consider the impact of social networks on users' interests. This paper proposes a new method by mining consumption information in social networks. First of all, a novel method based on a recurrent neural network is designed to express the correlation between users' purchasing behaviors, which dynamically captures users' interests in different scenarios. Secondly, to more accurately obtain the interests of neighbor nodes in social networks, this paper designs a multi-layer recurrent neural network with residual structure, which enriches the consumption information of neighbor nodes, and at the same time allows the original feature information of nodes to be retained during the propagation process of the deep network. Finally, the two interests are merged into users' potential purchase intention, which improves the performance of recommendation systems. The experimental results prove the effectiveness of the two methods proposed in this paper. Their Recall and Normalized Discounted Cumulative Gain on Douban Movie and Delicious datasets both outperform the state-of-art models.