| 摘 要: 针对意图标签数量过多带来的识别挑战,提出一种拆分策略,旨在降低训练的复杂度,并在解码层采用双头MLP(多层感知机)结合CRF(条件随机场)的机制,精准预测意图标签。实验结果表明,该方法的F1 值相较于基准模型提升至78%。此外,为验证大语言模型在意图识别任务上的表现,提出了一种面向医疗领域的大语言模型意图识别方法,其在处理多样化信息的数据时,能够深入探索并提炼出丰富的知识,将其与提出的深度学习方法进行对比,并深入分析二者在数据集上表现存在差异的原因,可为医疗领域意图识别的后续研究提供重要参考依据。 |
| 关键词: 意图识别;标签拆分;大语言模型;自然语言理解 |
|
中图分类号: TP391
文献标识码: A
|
| 基金项目: 教育部人文社会科学研究规划科学项目“融合深度神经网络模型的汉语隐喻计算研究”(18YJA740016);国家社会科学基金重大规划项目“汉语隐喻的逻辑表征与认知计算”(18ZDA290) |
|
| Research on Intent Recognition Methods in the Healthcare Field |
|
ZHANG Zhuoqun, WANG Rongbo, HUANG Xiaoxi
|
(Institute of Cognitive and Intelligent Computing, Hangzhou Dianzi University, Hangzhou 310018, China)
906976140@qq.com; wangrongbo@hdu.edu.cn; huangxx@hdu.edu.cn
|
| Abstract: In response to the recognition challenge caused by the excessive number of intent labels, this paper proposes a splitting strategy to reduce the training complexity. A dual-head MLP (Multi-Layer Perceptron) combined with a CRF (Conditional Random Field) mechanism in the decoding layer is employed to accurately predict intent labels. Experimental results show that this approach achieves an F1 value of 78% , significantly improving upon the baseline model. Furthermore, to evaluate the performance of large language models in intent recognition tasks, a large language model-based intent recognition method is proposed for the healthcare field. This method can deeply explore and extract rich knowledge when processing diverse information data. Finally, an experimental comparison is conducted between the proposed deep learning method and the large language model method to deeply analyze the differences in their performance on the data set and provide essential insights for future research in intent recognition within the healthcare field. |
| Keywords: intent recognition; label splitting; large language model; natural language understanding |
