Research on Multi-modal Optimization of GPT4.0 Based on User Feedback Data

-- Taking the Scenario of Medical Intelligent Customer Service as an Example

Abstract

In the medical intelligent customer service scenario, the accuracy of GPT-4.0's multimodal responses and user satisfaction still have room for improvement. This paper takes the medical intelligent customer service as the research scenario, focusing on the optimization of GPT-4.0's multimodal responses based on user feedback data. Firstly, the research background and issues are systematically elaborated, the current research status and gaps at home and abroad are sorted out, the significance and objectives of the research are clarified; secondly, a system for collecting, cleaning, annotating, and feature mining of user feedback data in medical intelligent customer service is constructed, providing a data foundation for optimization; thirdly, an optimization algorithm module containing feedback parsing, LoRA fine-tuning, and multimodal fusion is designed; finally, the optimization effect is verified through the establishment of an experimental environment, comparative experiments, and multi-index evaluation. The research results not only improve the response quality of GPT-4.0 in the medical intelligent customer service scenario but also provide theoretical and practical references for the optimization of multimodal large models in vertical fields.