Abstract:Chinese Named Entity Recognition (NER) is a very important subtask in the information extraction domain. Its purpose is to locate named entities in the text and classify them into predetermined categories. The key point of NER is to learn a high-quality representation of tokens. Recently, representation learning techniques have been introduced into NER due to their excellent performance in mining the semantics of texts and mastering their organization. In the field of Chinese, many studies introduce multimodal feature extraction schemes to enrich token representations, such as radical and word. However, the learning scheme of these auxiliary features is relatively complex and has difficulty learning the interactive relationship between features with a concatenation or MLP fusion method. To address these challenges, a Multimodal Chinese NER Model based on Self-attention Mechanism named MNER is proposed, which consists of a multimodal feature fusion module and an entity classification module. To study the informative characterization of tokens, a multimodal feature fusion module is proposed to exploit radical, character, and word information. In the multimodal feature fusion module, a self-attention mechanism is designed to integrate multimodal features based on the correlation between the features, which addresses the problem that existing methods have difficulty exploiting interaction information between features. A semantic-aware category modifier is proposed to enhance the CRF classification layer's performance. It increases entity discrimination by adjusting the embeddings of features according to the similarity between the token embeddings and embedding of each entity category, which widens the encoding gap between different entities and narrows the search scope for entity classification. Finally, the proposed MNER is compared with ten state-of-the-art methods on Weibo and Resume datasets, and the results show the superiority of our model on three metrics.

A Multi-task Approach for Machine Reading Comprehension Form Named Entity Recognition Tasks

Improving Biomedical Named Entity Recognition with a Unified Multi-Task MRC Framework

Named Entity Recognition via Machine Reading Comprehension: A Multi-Task Learning Approach

NER-to-MRC: Named-Entity Recognition Completely Solving as Machine Reading Comprehension

A Multi-task Framework for Named Entity Recognition

Active Learning for Name Entity Recognition with External Knowledge

Multi-Task Deep Neural Networks for Multi-Document Reading Comprehension

An Improved Method for Chinese Named Entity Recognition Based on MRC

Multi-Matching Network for Multiple Choice Reading Comprehension

A Unified MRC Framework for Named Entity Recognition

Query-Based Named Entity Recognition

Complex Named Entity Recognition via Deep Multi-task Learning from Scratch.

Named Entity Recognition Based on Multi-scale Attention

Multi-task Transformer with Relation-attention and Type-attention for Named Entity Recognition

Machine Reading Comprehension for Continuous Questions Via Multi-Task Learning

PMRC: Prompt-Based Machine Reading Comprehension for Few-Shot Named Entity Recognition

Frame-based Neural Network for Machine Reading Comprehension

Human-Machine Collaboration Based Named Entity Recognition.

Pretraining Multi-modal Representations for Chinese NER Task with Cross-Modality Attention

Multimodal Features Enhanced Named Entity Recognition Based on Self-Attention Mechanism.