计算机科学
对比度(视觉)
反问题
背景(考古学)
反向
算法
人工智能
数学
物理
动量(技术分析)
期限(时间)
透视图(图形)
牙石(牙科)
光学(聚焦)
噪音(视频)
钥匙(锁)
作者
Di Wu,Fanming Meng,Yao Peng
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2026-06-01
卷期号:: 1-7
标识
DOI:10.1109/ijcnn69854.2026.11729393
摘要
Dialogue generation is an essential component of task-oriented dialogue systems, aiming to assist users in completing domain-specific tasks via natural language interaction. Existing few-shot dialogue generation models face challenges in perceiving domain dialogue state and generating responses that are inconsistent with the dialogue context. To address these challenges, we propose the IPCMC-TOD model, a Few-shot Dialogue Generation Model with Inverse Prompt Response Reconstruction and Contextual Momentum Contrast. Specifically, considering the critical role of the model's adaptability to domain dialogue state, we design an inverse prompt response reconstruction. An inverse-guided dual-prompt template is constructed. An inverse-prompt-trained slot extractor is employed to reconstruct the original system responses. To enhance the logical consistency of dialogue responses, we design a Contextual Momentum Contrast approach. Based on the posterior inference response, the complete set of samples is categorised into positive and negative sets. Momentum-based pseudo-samples are generated by the momentum encoder. We define a contrast loss function with margin constraints for positive and negative samples. This function significantly enhances the fine-grained discrimination ability of the response selector. Experimental results on the MultiWOZ 2.0 and MultiWOZ 2.1 datasets demonstrate the effectiveness of IPCMC-TOD over all baseline models.
科研通智能强力驱动
Strongly Powered by AbleSci AI