Few-shot Learning for Multi-label Intent Detection
This addresses the problem of detecting multiple user intents with limited labeled data, which is incremental as it builds on existing thresholding and embedding methods.
The paper tackles few-shot multi-label intent detection by learning universal thresholds from data-rich domains and adapting them to few-shot domains with nonparametric calibration, while using label name embeddings to improve relevance scores. Experiments on two datasets show significant outperformance over baselines in one-shot and five-shot settings.
In this paper, we study the few-shot multi-label classification for user intent detection. For multi-label intent detection, state-of-the-art work estimates label-instance relevance scores and uses a threshold to select multiple associated intent labels. To determine appropriate thresholds with only a few examples, we first learn universal thresholding experience on data-rich domains, and then adapt the thresholds to certain few-shot domains with a calibration based on nonparametric learning. For better calculation of label-instance relevance score, we introduce label name embedding as anchor points in representation space, which refines representations of different classes to be well-separated from each other. Experiments on two datasets show that the proposed model significantly outperforms strong baselines in both one-shot and five-shot settings.