Multi-Aspect co-Attentional Collaborative Filtering for Extreme Multi-label Text Classification

Anonymous

Abstract: This work proposes a general and effective architecture for the extreme multi-label text classification (XMTC), and reformate the learning task to an interaction function between document and label. Recently, there are many studies trying to enhance text representation or reduce the number of labels to optimize the problem of lack of information in a text or the sparsity of the possibility vector. In the field of recommendation, a similar problem is already defined and studied for a quite long time. It is worthy to learn methods from recommendation to XMTC for finding matching relations in large size of dataset accurately. With co- attention mechanism and neural collaborative filtering, we not only learn informative label representation enhanced by document-specific label group vector and label-specific text feature vector but also build an effective interaction function to get matching score. After ex- tensive comparison experiments with various models, results demonstrate the architecture we proposed outperforms most of the methods and achieves significant improvement on basic document encoders.

Paper Type: long