摘要
Abstract
This study addresses the issues of increasing medical complaints,slow response times from customer service,and inadequate emotional understanding,by constructing a multimodal emotion recognition and complaint handling system for medical scenarios.The system integrates the WavLM and BERT models to extract speech and text features,respectively,and employs a hybrid mechanism to distinguish among three types of emotions and generate response strategies.Experiments demonstrate that the multimodal model achieves an emotion recognition accuracy rate of nearly 90%,with precision,recall,and F1 scores all exceeding 86%.This represents a significant improvement compared to the single speech modality(with indicators ranging from 74%to 79%)and the single text modality(with indicators ranging from 78%to 82%),with an average improvement of approximately 10 percentage points across the four core indicators.Actual deployment tests have verified that the system exhibits good recognition accuracy and response efficiency,making it practical and worthy of promotion.关键词
医疗服务/多模态情绪识别/深度学习/智能客服Key words
Medical Service/Multimodal Emotion Recognition/Deep Learning/Intelligent Customer Service分类
医药卫生
