NameRec*: Highly Accurate and Fine-grained Person Name Recognition

Person names are essential entities in the Named Entity Recognition (NER) task. Traditional NER models have good performance in recognising well-formed person names from text with consistent and complete syntax, such as news articles. However, user-generated documents such as academic homepages and articles in online forums may contain lots of free-form text with incomplete syntax and person names in various forms. To address person name recognition in this context, we propose a fine-grained annotation scheme based on anthroponymy. To take full advantage of the fine-grained annotations, we propose a Co-guided Neural Network (CogNN) for person name recognition. CogNN fully explore the intra-sentence context and rich training signals of name forms. However, the inter-sentence context and implicit relations, which are extremely essential for recognizing person names in long documents, are not captured. To address this issue, we propose a Multi-inference Overlapped BERT Model (NameRec*) through an overlapped input processor, and an inter-sentence encoder with bidirectional overlapped contextual embedding learning and multiple inference mechanisms. NameRec* takes full advantage of inter-sentence context in long documents, while loses advantage for short documents without too much inter-sentence context. To derive benefit from different documents with diverse abundance of context, we further propose an advanced Adaptive Multi-inference Overlapping BERT Model (Ada-NameRec*) to dynamically adjust the inter-sentence overlapping ratio to different documents. We conduct extensive experiments to demonstrate the superiority of the proposed methods on both academic homepages and news articles.