基于Prompt的两阶段澄清问题生成方法  

Two-stage clarification question generation method based on Prompt

作　　者：王培冰 张宁[1] 张春[1] Wang Peibing;Zhang Ning;Zhang Chun(School of Computer&Information Technology,Beijing Jiaotong University,Beijing 100044,China)

机构地区：[1]北京交通大学计算机与信息技术学院,北京100044

出　　处：《计算机应用研究》2024年第2期421-425,共5页Application Research of Computers

基　　金：国家重点研发计划资助项目(2019YFB1405202)。

摘　　要：在自然语言相关系统中,当用户输入存在歧义时,生成澄清问题询问用户有助于系统理解用户需求;基于Prompt的方法可以更好地挖掘预训练语言模型的潜在知识,但往往需要手动设计模板,限制其生成澄清问题的多样性。为解决这一问题,提出了TSCQG(two-stage clarification question generation)方法。首先,在动态Prompt模板生成阶段,利用歧义上下文和预训练语言模型生成动态的Prompt模板;然后在缺失信息生成阶段,将Prompt模板与外部知识相结合,充分利用预训练语言模型的生成能力生成相应的缺失信息。实验结果表明,在CLAQUA数据集的多轮对话情况中,BLEU值和ROUGE-L值分别达到了58.31和84.33,在ClariQ-FKw数据集上,BLEU值和ROUGE-L值分别达到了31.18和58.86。实验结果证明了TSCQG方法在澄清问题生成任务上的有效性。In natural language-oriented systems,generating clarification questions to ask users when their input is ambiguous can help the system better understand the user’s requirements.Although Prompt-based approaches can better exploit the latent knowledge of pre-trained language models,they often require hand-designed templates,constraining their diversity in generating clarification questions.To address this limitation,this paper proposed the two-stage clarification question generation(TSCQG)method.Firstly,in the dynamic Prompt template generation stage,the TSCQG method used the ambiguous context and the pre-trained language models to generate Prompt templates.Then,in the missing information generation stage,it combined the Prompt templates and relevant external knowledge and capitalized on the generative potential of the pre-trained model to gene-rate relevant missing information.Experimental results demonstrate that the BLEU value and ROUGE-L value of the multi-round dialogue situation on the CLAQUA dataset reach 58.31 and 84.33,and the BLEU value and ROUGE-L value on the ClariQ-FKw dataset reach 31.18 and 58.86,respectively.The experimental results validate the effectiveness of the TSCQG method in clarification question generation tasks.

关 键 词：预训练语言模型 PROMPT 澄清问题生成 自然语言系统 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]