AttentionXML: Label Tree-based Attention-Aware Deep Model for High-Performance Extreme Multi-Label Text Classification

Extreme multi-label text classification (XMTC) is an important problem in the of big data, for tagging a given text with the most relevant multiple from an extremely large-scale label set. XMTC can be found in many, such as item categorization, web page tagging, and news. Traditionally most methods used bag-of-words (BOW) as inputs, word context as well as deep semantic information. Recent attempts to the problems of BOW by deep learning still suffer from 1) failing to the important subtext for each label and 2) lack of scalability against huge number of labels. We propose a new label tree-based deep learning for XMTC, called AttentionXML, with two unique features: 1) a multi-label mechanism with raw text as input, which allows to capture the most part of text to each label; and 2) a shallow and wide probabilistic tree (PLT), which allows to handle millions of labels, especially for"tail labels". We empirically compared the performance of AttentionXML with of eight state-of-the-art methods over six benchmark datasets, including-3M with around 3 million labels. AttentionXML outperformed all competing under all experimental settings. Experimental results also show that achieved the best performance against tail labels among label-based methods. The code and datasets are available at://github.com/yourh/AttentionXML .