Word-class embeddings for multiclass text classification

Pre-trained word embeddings encode general semantics and lexical regularities of natural language, have proven useful across many NLP tasks, including sense disambiguation, machine translation, sentiment analysis, to name a few. In supervised tasks such as multiclass text classification (the focus this article) it seems appealing enhance representations with ad-hoc that task-specific information. We propose (supervised) word-class (WCEs), show that, when concatenated (unsupervised) pre-trained embeddings, they substantially facilitate the training deep-learning models in by topic. empirical evidence WCEs yield consistent improvement accuracy, using six popular neural architectures widely used publicly available datasets for classification. One further advantage method is conceptually simple straightforward implement. Our code implements at https://github.com/AlexMoreo/word-class-embeddings .