Auxiliary Loss for BERT-Based Paragraph Segmentation

Paragraph segmentation is a text task. Iikura et al. achieved excellent results on paragraph by introducing focal loss to Bidirectional Encoder Representations from Transformers. In this study, we investigated Daily News and Novel datasets. Based the approach proposed al., used auxiliary train model improve performance. Consequently, average F1-score obtained of was 0.6704 dataset, whereas that our 0.6801. Our thus improved performance approximately 1%. The improvement also confirmed dataset. Furthermore, two-tailed paired t-tests indicated there statistical significance between two approaches.