Research on the Application System of Big Models Based on Deep Learning in the Field of Financial Customer Service

This paper proposes research on the application system of big models based on deep learning in the field of financial customer service. In view of the problems of insufficient understanding of complex financial contexts and inaccurate responses of existing financial customer service systems, this paper designs an adaptive financial dialogue generation algorithm. The article uses the pre-trained model qwen-2.5 instead of GPT-3, and the comparison models are changed to GPT-3.5 and other mainstream training models. By introducing special vocabulary in the financial field for adjustment, combined with adaptive attention mechanism and financial entity embedding, it enhances the semantic understanding ability of financial scenarios and the dynamic capture ability of customer intentions. The system simulation experiment uses a real dialogue dataset of a financial enterprise, which contains 100,000 annotated dialogue data. The results show that compared with traditional dialogue generation models (the comparison models are changed to GPT-3.5 and other mainstream training models), the model proposed in this paper improves the response accuracy by 12.3 % and reduces the response time by 18.6 %. In addition, in the user satisfaction survey, the customer satisfaction of using this system increased by 22 %. Data experiments also show that this system can handle complex financial problems more accurately, such as multiaccount transaction record query and personalized investment advice generation. This study proves the practical application value of large models based on deep learning in the field of financial customer service, which not only improves customer service efficiency, but also enhances user experience, and provides new ideas for the development of intelligent customer service systems.