| 摘 要: 针对员工情感文本识别中样本数量有限且情感表达复杂的问题,提出结合常识知识模型CoMet与预训练语言模型LERT的情感识别方法。首先,利用DeepSeek生成涵盖悲伤、信任、喜悦、恐惧、怀疑、愤怒和自我否定7类情感的员工对话数据集,并进行情感倾向标注;然后,基于LERT模型结合CoMet模型,通过引入对话主体的目标识别与常识知识,增强模型对情感状态的理解与识别能力。实验结果表明,该方法在情感识别准确性上较传统方法平均提升10.2%,为情感分析提供了新的技术路径。 |
| 关键词: 深度学习 LERT CoMet 对话情感识别 |
|
中图分类号:
文献标识码: A
|
| 基金项目: 教育部人文社会科学研究一般项目(23YJCZH281);上海市哲学社会科学规划课题(2022ZGL010);信息网络安全公安部重点实验室开放课题 |
|
| Research on Generative Employee Emotion Recognition Based on LERT and CoMet |
|
WANG Tingxuan, YIN Pei
|
(Business School, University of Shanghai for Science & Technology, Shanghai 200093, China)
15968612285@163.com; pyin@usst.edu.cn
|
| Abstract: To address the challenges of limited sample size and complex emotional expressions in employee sentiment text recognition, this paper proposes an emotion recognition method combining the commonsense knowledge model CoMet and the pre-trained language model LERT. First, a dataset of employee dialogues covering seven emotion categories (sadness, trust, joy, fear, doubt, anger, and sel-f denial) was generated using DeepSeek, with sentiment annotations applied. Subsequently, by integrating CoMet with the LERT model, the approach enhances the model’s
understanding and recognition of emotional states through intent recognition of dialogue subjects and commonsense knowledge integration. Experimental results demonstrate that this method achieves an average improvement of 10.2 percentage points in emotion recognition accuracy compared to traditional approaches, offering a novel technical pathway for sentiment analysis. |
| Keywords: deep learning LERT CoMet conversational emotion recognition |
