On Robustness and Bias Analysis of BERT-Based Relation Extraction

Fine-tuning pre-trained models have achieved impressive performance on standard natural language processing benchmarks. However, the resultant model generalizability remains poorly understood. We do not know, for example, how excellent can lead to perfection of generalization models. In this study, we analyze a fine-tuned BERT from different perspectives using relation extraction. also characterize differences in techniques according our proposed improvements. From empirical experimentation, find that suffers bottleneck terms robustness by way randomizations, adversarial and counterfactual tests, biases (i.e., selection semantic). These findings highlight opportunities future Our open-sourced testbed DiagnoseRE is available https://github.com/zjunlp/DiagnoseRE .