Gaussian Prior Reinforcement Learning for Nested Named Entity Recognition

TL;DR

This paper introduces GPRL, a novel reinforcement learning-based seq2seq model for nested NER that models entity recognition as triplet sequence generation, effectively capturing boundary relations and improving performance.

Contribution

The paper proposes GPRL, a new approach that uses Gaussian priors and reinforcement learning to better model nested entity boundaries and recognition order in NER tasks.

Findings

GPRL outperforms previous nested NER models on three datasets.

The Gaussian prior effectively models boundary distance distributions.

Reinforcement learning helps decouple entity order from gold labels.

Abstract

Named Entity Recognition (NER) is a well and widely studied task in natural language processing. Recently, the nested NER has attracted more attention since its practicality and difficulty. Existing works for nested NER ignore the recognition order and boundary position relation of nested entities. To address these issues, we propose a novel seq2seq model named GPRL, which formulates the nested NER task as an entity triplet sequence generation process. GPRL adopts the reinforcement learning method to generate entity triplets decoupling the entity order in gold labels and expects to learn a reasonable recognition order of entities via trial and error. Based on statistics of boundary distance for nested entities, GPRL designs a Gaussian prior to represent the boundary distance distribution between nested entities and adjust the output probability distribution of nested boundary tokens. Experiments on three nested NER datasets demonstrate that GPRL outperforms previous nested NER models.