Research on Personalized Financial Product Recommendation by Integrating Large Language Models and Graph Neural Networks
Y. X. Zhao, Yike Peng, Dannier Li, Yuxin Yang, Chengrui Zhou, Jing Dong
Abstract
As fintech expands fast, the relevance of personalization of financial products recommendations has gained ground. Latent preferences and complicated relationships between users cannot be modelled with the traditional methods such as collaborative filtering or content-based models. Our hypothesis is a hybrid framework that combines large language models (LLMs) with graph neural networks (GNNs). A pre-trained LLM can embed text data (e.g., user reviews) into dense feature vectors, and a heterogeneous user -product graph can capture interactions and social connections. The GNN combines text and graph information by passing a custom message-passing mechanism that is jointly optimized. Our model achieves state-of-the-art results compared to standalone LLM or GNN on experiments run on public and real-world financial datasets, and is highly interpretable. The work provides two novelties in the direction of personalized financial advice and cross-modal fusion in recommender tasks in general.