Unknown Intent Detection Using Gaussian Mixture Model with an Application to Zero-shot Intent Classification

• DOI: 10.18653/v1/2020.acl-main.99
• Publication Date: 2020-07-29T14:14:43Z
• Authors: Guangfeng Yan; Lu Fan; Qimai Li; Han Liu; Xiaotong Zhang; Xiao-Ming Wu; Albert Y.S. Lam
• Keywords: One-class classification; Utterance; Contextual image classification

User intent classification plays a vital role in dialogue systems. Since user may frequently change over time many realistic scenarios, unknown (new) detection has become an essential problem, where the study just begun. This paper proposes semantic-enhanced Gaussian mixture model (SEG) for detection. In particular, we utterance embeddings with distribution and inject dynamic class semantic information into means, which enables learning more class-concentrated that help to facilitate downstream outlier Coupled density-based algorithm, SEG achieves competitive results on three real task-oriented datasets two languages On top of that, propose integrate as identifier existing generalized zero-shot models improve their performance. A case state-of-the-art method, ReCapsNet, shows can push performance significantly higher level.