TLC-XML: Transformer with Label Correlation for Extreme Multi-label Text Classification

• DOI: 10.1007/s11063-024-11460-z
• Publication Date: 2024-02-10T11:02:10Z
• Authors: Fei Zhao; Qing Ai; Xiangna Li; Wenhui Wang; Qingyun Gao; Yichun Liu
• Keywords: 0202 electrical engineering, electronic engineering, information engineering; 02 engineering and technology

Abstract Extreme multi-label text classification (XMTC) annotates related labels for unknown from large-scale label sets. Transformer-based methods have become the dominant approach solving XMTC task due to their effective representation capabilities. However, existing fail effectively exploit correlation between in task. To address this shortcoming, we propose a novel model called TLC-XML, i.e., Transformer with extreme classification. TLC-XML comprises three modules: Partition, Matcher and Ranker. In Partition module, semantic co-occurrence information of construct graph, further partition strongly correlated into same cluster. cluster learning, which uses graph convolutional network (GCN) extract clusters. We then introduce these valuable correlations classifier match Ranker interaction aggregates raw prediction neighboring labels. The experimental results on benchmark datasets show that significantly outperforms state-of-the-art methods.