Multi-Label Text Classification Based on BERT and Label Attention Mechanism
Xinghong Chen, Yifeng Yin, Tao Feng · 2023
In natural language processing, multi-label text classification is a crucial task. Recently, many methods had introduced information related to labels, which had improved the classification effect of methods. This paper proposed a model that utilizes BERT and a label attention mechanism to effectively leverage the semantic information present in labels. Through fine-tuning of BERT, the textual and label data were transformed into vector representations, and then the label attention mechanism was employed to extract text features, which are more relevant to labels. Finally, a corresponding classifier was constructed to complete the classification task. Experiments show that compared with the baselines mentioned in the paper, the proposed method had an improvement on both AAPD and RCVl-v2 datasets, which proved the effectiveness of the proposed method in the multi-label text classification task.