Attention-Based Sub-Word Network for Multilingual Short Text Classification

Abstract

Feature computation of multilingual text is an important semantic processing task in the field of natural language processing (NLP). In the actual production environment, the state-of-the-art models cannot analyze short texts mixed with multi-languages correctly. To tackle these problems, we propose a sub-word embedding network with multilingual features for short text understanding to capture the most important semantic information in a multilingual short sentence. In this work, our method utilizes a coupling coefficient calculation-based model that generates the sub-words of the input sentence. By sharing sub-word features, the feature space of multilingual mixed-word is constructed. The method that can extract the most significant information in a sentence without ignoring other relevant information. While the model is structurally simple, it can easily be trained end-to-end and scales to a large amount of training data. The experimental results on the Multilingual Short Text (MST), THUCNews and AGNews datasets show that our method outperforms most of the existing methods.