A Neural Influence Diffusion Model for Social Recommendation

Abstract

Precise user and item embedding learning is the key to building a successful recommender system. Traditionally, Collaborative Filtering (CF) provides a way to learn user and item embeddings from the user-item interaction history. However, the performance is limited due to the sparseness of user behavior data. With the emergence of online social networks, social recommender systems have been proposed to utilize each user's local neighbors' preferences to alleviate the data sparsity for better user embedding modeling. We argue that, for each user of a social platform, her potential embedding is influenced by her trusted users, with these trusted users are influenced by the trusted users' social connections. As social influence recursively propagates and diffuses in the social network, each user's interests change in the recursive process. Nevertheless, the current social recommendation models simply developed static models by leveraging the local neighbors of each user without simulating the recursive diffusion in the global social network, leading to suboptimal recommendation performance. In this paper, we propose a deep influence propagation model to stimulate how users are influenced by the recursive social diffusion process for social recommendation. For each user, the diffusion process starts with an initial embedding that fuses the related features and a free user latent vector that captures the latent behavior preference. The key idea of our proposed model is that we design a layer-wise influence propagation structure to model how users' latent embeddings evolve as the social diffusion process continues. We further show that our proposed model is general and could be applied when the user (item) attributes or the social network structure is not available. Finally, extensive experimental results on two real-world datasets clearly show the effectiveness of our proposed model 1 , with more than 13% performance improvements over the best baselines for top-10 recommendation on the two datasets.

Graph convolutional networks: a comprehensive review

“…Graph convolutional networks have been widely used for social recommendation which aims to leverage the user-item interactions and/or user-user interactions to boost the recommendation performance. Wu et al propose a neural influence diffusion model that takes how users are influenced by their trusted friends into considerations for better social recommendations [124]. Ying et al propose a very efficient graph convolutional network model PinSage [125] based on GraphSAGE [61] which exploits the interactions between pins and boards in Pinterest.…”

A Heterogeneous Graph Neural Model for Cold-start Recommendation

“…We compare our HGNR model with the following baselines: BPR [11], NeuMF [1], GC-MC [12], DiffNet [14] and NGCF [13]. Our experiments have two parts: Firstly we examine the social sub-graph and the review sub-graph separately, so that we can clearly demonstrate the effectiveness of each sub-graph; Secondly we compare our HGNR model with all baselines in terms of Normalised Discounted Cumulative Gain@10 (NDCG) and Hit Ratio@10 (HR).…”

“…Furthermore, Wang et al [13] proposed the NGCF model to enhance performance by stacking three GCN layers. Wu et al proposed Diffnet [14], which simulates the social influence by applying high-order GCNs to learn from the social network graph. However, Diffnet solely used any existing social links to modify the users' embeddings, which means that the items' embeddings cannot benefit from the GCNs.…”

“…The social sub-graph S is a user-by-user square matrix initially filled by any existing social links between users. We use the existing social links to enhance the user embedding generation because one user's preferences can be affected by other users through social influence [6,14]. However, in the most widely used item recommendation datasets, the social graph is usually sparse, since most of the users are not included in the social graph [16].…”

Generating Reliable Friends via Adversarial Training to Improve Social Recommendation

“…The instance in this section just sets an example for other applications. Finally, there is an for each user u do 5 Input the seeded friends S u into the generator; 6 Generates p u over the whole user set; 7 Feed p u into the first Gumbel-Softmax layer; 8 Get a one-hot vector representing the friend v; Look-up operation for the items consumed by this friend;…”