User Cold-Start Recommendation via Inductive Heterogeneous Graph Neural Network

Desheng Cai,Shengsheng Qian,Quan Fang,Jun Hu,Changsheng Xu

DOI: https://doi.org/10.1145/3560487

IF: 4.657

2023-02-07

ACM Transactions on Information Systems

Abstract:Recently, user cold-start recommendations have attracted a lot of attention from industry and academia. In user cold-start recommendation systems, the user attribute information is often used by existing approaches to learn user preferences due to the unavailability of user action data. However, most existing recommendation methods often ignore the sparsity of user attributes in cold-start recommendation systems. To tackle this limitation, this article proposes a novel Inductive Heterogeneous Graph Neural Network (IHGNN) model, which utilizes the relational information in user cold-start recommendation systems to alleviate the sparsity of user attributes. Our model converts new users, items, and associated multimodal information into a Modality-aware Heterogeneous Graph (M-HG) that preserves the rich and heterogeneous relationship information among them. Specifically, to utilize rich and heterogeneous relational information in an M-HG for enriching the sparse attribute information of new users, we design a strategy based on random walk operations to collect associated neighbors of new users by multiple times sampling operation. Then, a well-designed multiple hierarchical attention aggregation model consisting of the intra- and inter-type attention aggregating module is proposed, focusing on useful connected neighbors and neglecting meaningless and noisy connected neighbors to generate high-quality representations for user cold-start recommendations. Experimental results on three real datasets demonstrate that the IHGNN outperforms the state-of-the-art baselines.

computer science, information systems

What problem does this paper attempt to address?

This paper aims to address the user cold-start recommendation problem. In traditional recommendation systems, for new users or new items, due to the lack of historical behavior data, existing methods usually rely on user attribute information to infer their preferences. However, most existing recommendation methods ignore the sparsity issue of user attributes in cold-start recommendation systems. To overcome this limitation, this paper proposes a new Inductive Heterogeneous Graph Neural Network (IHGNN) model.

IHGNN leverages relational information in user cold-start recommendation systems to alleviate the sparsity of user attributes. Specifically, the model transforms new users, items, and their related multimodal information into a Modality-aware Heterogeneous Graph (M-HG), which retains the rich and heterogeneous relational information between them. To utilize the rich heterogeneous relational information in M-HG to enrich the sparse attribute information of new users, the researchers designed a strategy based on random walk operations, collecting associated neighbors of new users through multiple sampling operations. Additionally, a multi-level attention aggregation model is proposed, including intra-type attention aggregation modules and inter-type attention aggregation modules, focusing on useful connected neighbors and ignoring meaningless and noisy neighbors to generate high-quality new user representations.

Experimental results show that IHGNN outperforms existing state-of-the-art baseline models on three real-world datasets. In summary, this paper mainly addresses the problem of how to effectively model rich heterogeneous relational information and generate high-quality new user representations in cold-start recommendation scenarios.