Label-Specific Document Representation for Multi-Label Text Classification

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 466–475, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 466 Label-Specific Document Representation for Multi-Label Text Classification Lin Xiao, Xin Huang, Boli Chen, Liping Jing Beijing Key Lab of Traffic Data Analysis and Mining Beijing Jiaotong University, China {17112079, 18120367, 18120345, lpjing}@bjtu.edu.cn Abstract Multi-label text classification (MLTC) aims to tag most relevant labels for the given docu- ment. In this paper, we propose a Label- Specific Attention Network (LSAN) to learn the new document representation. LSAN takes advantage of label semantic information to determine the semantic connection between labels and document for constructing label- specific document representation. Meanwhile, the self-attention mechanism is adopted to i- dentify the label-specific document represen- tation from document content information. In order to seamlessly integrate the above two parts, an adaptive fusion strategy is designed, which can effectively output the comprehen- sive document representation to build multi- label text classifier. Extensive experimental re- sults on four benchmark datasets demonstrate that LSAN consistently outperforms the state- of-the-art methods, especially on the predic- tion of low-frequency labels. The code and hyper-parameter settings are released to facili- tate other researchers 1. 1 Introduction Text classification is a fundamental text mining task including multi-class classification and multi- label classification. The former only assigns one label to the given document, while the latter clas- sifies one document into different topics. In this paper, we focus on multi-label text classification (MLTC) because it has become one of the core tasks in natural language processing and has been widely applied in topic recognition (Yang et al., 2016), question answering (Kumar et al., 2016), sentimental analysis (Cambria et al., 2014) and so on. With the boom of big data, MLTC becomes significantly challenging because it has to handle 1https://github.com/EMNLP2019LSAN/ LSAN/ the massive documents, words and labels simulta- neously. Therefore, it is an emergency to devel- op effective multi-label text classifier for various practical applications. Multi-label text classification allows for the co- existence of more than one label in a single doc- ument, thus, there are semantical correlations a- mong labels because they may share the same subsets of document. Meanwhile, the document may be long and complicated semantic informa- tion may be hidden in the noisy or redundant con- tent. Furthermore, most documents fall into few labels while a large number of “tail labels” on- ly contain very few positive documents. To han- dle these issues, researchers pay much attention on three facets: 1) how to sufficiently capture the semantic patterns from the original documents, 2) how to extract the discriminative information re- lated to the corresponding labels from each docu- ment, and 3) how to accurately mine the correla- tion among labels. Till now, in the community of machine learn- ing and natural language processing, researcher- s have paid tremendous efforts on developing MLTC methods in each facet. Among them, deep learning-based methods such as CNN …

Content truncated to 500 words