Towards Textual Out-of-Domain Detection without In-Domain Labels
This addresses a challenging real-world scenario for machine learning models, such as in intent classification, where label scarcity is common, offering a practical solution for OOD detection without requiring labeled in-domain data.
The paper tackles the problem of detecting out-of-domain (OOD) text inputs without access to in-domain labels, proposing a novel method that combines unsupervised clustering and contrastive learning to learn better data representations. The method significantly outperforms likelihood-based approaches and is competitive with state-of-the-art supervised methods, as demonstrated through extensive experiments.
In many real-world settings, machine learning models need to identify user inputs that are out-of-domain (OOD) so as to avoid performing wrong actions. This work focuses on a challenging case of OOD detection, where no labels for in-domain data are accessible (e.g., no intent labels for the intent classification task). To this end, we first evaluate different language model based approaches that predict likelihood for a sequence of tokens. Furthermore, we propose a novel representation learning based method by combining unsupervised clustering and contrastive learning so that better data representations for OOD detection can be learned. Through extensive experiments, we demonstrate that this method can significantly outperform likelihood-based methods and can be even competitive to the state-of-the-art supervised approaches with label information.