Research on Investor Education in China Based on Large Language Models

This study addresses the issue of investor education in the context of advisory marketing within China's wealth management industry. It proposes a method for profiling investors psychologically based on large language models (LLMs) and the “Id-Ego-Superego” psychological framework. This approach utilizes data from advisory marketing dialogues enhanced by GPT-4o. The Qwen 2.5 70b model was fine-tuned to create a labeling system capable of identifying the dominant psychological type (Id, Ego, or Superego) of clients. The results show that this system can accurately identify clients' psychological characteristics (with an average accuracy of 91.5%) and significantly improve the effectiveness of investor education, as reflected in the increased success rate of investment recommendations, reduced client decision-making time, and improved client satisfaction. Compared with traditional rule-based and SVM-based models, this model demonstrates significant improvements across all metrics. This study emphasizes the concept of “people-oriented” investor education and highlights the potential of LLMs in understanding investor psychology and promoting personalized education.