Cold-Start Product Recommendation Method Based on GAE

Abstract

Recommendation systems play a pivotal role in alleviating information overload, delivering customized advisory services, and aiding users in making investment decisions. However, the cold start problem in recommendation systems has always been in urgent need of solution and optimization, especially for new item uptake or new user engagement. Owing to this, this paper categorizes the conventional and cutting-edge methodologies for addressing the cold start problem, and elucidates the research progress and outstanding methods in recent years. Firstly, this paper provides an exhaustive review of the various strategies and methods proposed by researchers to alleviate the cold start problem. Secondly, the paper summarizes several traditional recommendation methods for mitigating cold-start recommendations. Thirdly, the paper synthesizes the recent strategies and approaches into two categories: data-driven strategies and approach-driven strategies. Furthermore, the approach-driven strategies are categorized into five main clusters based on meta learning, session, heterogeneous graph and attributed graph, and other novel approaches, Furthermore, the method-driven strategies are refined into five primary clusters: meta-learning, session-based learning, heterogeneous graphs, attributed graphs, and other innovative methods. Particularly, a model named GAE, composed of graph attention neural networks (GAT) and meta-learning, is highlighted. This model initially aggregates and extracts attribute feature information from the neighbor nodes of cold-start users using the GAT; then, it utilizes the Embedding Generator (EG) to consider the cold-start users' own attribute information, combining it with the associated neighbor attribute feature information to generate the final attribute feature information for the cold-start users; meta-learning is employed to generate a list of recommended items for the cold-start users. Finally, the possible research directions to solve the cold start problem in the future are pointed out, which include complex relationship learning, multimodal recommendation, and privacy protection.