Fine-grained Chinese Named Entity Recognition in Entertainment News Using Adversarial Multi-task Learning

Abstract

Named Entity Recognition (NER) systems are often used as the first step for most of Natural Language Processing (NLP) tasks like question answering, information retrieval, etc. In our paper, the fine-grained NER for entertainment news is the task of identifying nine named entities like person, TV, music, etc. It is challenging not only because of inherent noisiness but the data contains numerous informal abbreviations. We proposed a novel adversarial multi-task learning framework by employing two auxiliary tasks named Part-of-Speech (POS) and Chinese Word Segmentation (CWS) together with a primary NER task. Our model can use task-shared features better and filter task-specific information of CWS and POS. To the best of our knowledge, our work is the first study that incorporate POS and CWS into fine-grained NER task in entertainment news.