Domain Intent
Domain intent, specifically the detection and classification of user utterances falling outside a system's predefined domain (out-of-domain or OOD intents), is a crucial area in natural language understanding for task-oriented dialogue systems. Current research focuses on improving OOD detection accuracy using various techniques, including dual encoders, large language models (LLMs), and prototype-based learning, often incorporating strategies like contrastive learning and soft labeling to address challenges such as data imbalance and noisy pseudo-labels. These advancements aim to enhance the robustness and adaptability of dialogue systems by enabling them to handle unexpected user inputs and dynamically expand their capabilities. The resulting improvements in OOD detection have significant implications for building more resilient and user-friendly conversational AI systems.
Papers
Watch the Neighbors: A Unified K-Nearest Neighbor Contrastive Learning Framework for OOD Intent Discovery
Yutao Mou, Keqing He, Pei Wang, Yanan Wu, Jingang Wang, Wei Wu, Weiran Xu
Disentangling Confidence Score Distribution for Out-of-Domain Intent Detection with Energy-Based Learning
Yanan Wu, Zhiyuan Zeng, Keqing He, Yutao Mou, Pei Wang, Yuanmeng Yan, Weiran Xu