Are Pretrained Transformers Robust in Intent Classification? A Missing Ingredient in Evaluation of Out-of-Scope Intent Detection
This work addresses robustness issues in intent detection for NLP applications, highlighting a critical evaluation gap.
The paper identifies that pre-trained Transformer models are vulnerable to in-domain but out-of-scope (ID-OOS) examples in few-shot intent classification, showing poor performance on new datasets constructed for this purpose.
Pre-trained Transformer-based models were reported to be robust in intent classification. In this work, we first point out the importance of in-domain out-of-scope detection in few-shot intent recognition tasks and then illustrate the vulnerability of pre-trained Transformer-based models against samples that are in-domain but out-of-scope (ID-OOS). We construct two new datasets, and empirically show that pre-trained models do not perform well on both ID-OOS examples and general out-of-scope examples, especially on fine-grained few-shot intent detection tasks. To figure out how the models mistakenly classify ID-OOS intents as in-scope intents, we further conduct analysis on confidence scores and the overlapping keywords, as well as point out several prospective directions for future work. Resources are available on https://github.com/jianguoz/Few-Shot-Intent-Detection.