Named Entity Recognition Networks Based on Syntactically Constrained Attention

The task of named entity recognition can be transformed into a machine reading comprehension by associating the query and its context, which contains information, with encoding layer. In this process, model learns priori knowledge about entity, from query, to achieve good results. However, as length context increases, struggles an increasing number less relevant words, distract it task. Although attention mechanisms help understand contextual semantic relations, without explicit constraint may allocated task-relevant leading bias in model’s understanding context. To address problem, we propose new model, syntactic constraint-based dual-context aggregation network, uses information guide modeling. By incorporating mechanism, better determine relevance each word task, selectively focus on parts This enhances ability read ultimately improving performance tasks. Extensive experiments three datasets, ACE2004, ACE2005, GENIA, show that method achieves superior when compared previous methods.