Confidence Calibration for Intent Detection via Hyperspherical Space and Rebalanced Accuracy-Uncertainty Loss

Data-driven methods have achieved notable performance on intent detection, which is a task to comprehend user queries. Nonetheless, they are controversial for over-confident predictions. In some scenarios, users do not only care about the accuracy but also confidence of model. Unfortunately, mainstream neural networks poorly calibrated, with large gap between and confidence. To handle this problem defined as calibration, we propose model using hyperspherical space rebalanced accuracy-uncertainty loss. Specifically, project label vector onto uniformly generate dense representation matrix, mitigates predictions due overfitting sparse one-hot matrix. Besides, rebalance samples different uncertainty better guide training. Experiments open datasets verify that our outperforms existing calibration achieves significant improvement metric.