Causality-aware Deconfounded NER Network for Chinese Named Entity Recognition

ACL ARR 2026 January Submission240 Authors

Keywords: Chinese Named Entity Recognition, Causal Deconfounding, Multi-granularity Feature Extraction, Prototype Contrastive Learning, Complex Text Processing

Abstract: Chinese Named Entity Recognition (Chinese NER) faces challenges such as ambiguous entity boundaries and limited classification accuracy, mainly due to the lack of clear word boundaries and the strong coupling between semantic features and interfering features. To address the interference caused by confounding factors like domain bias, this paper proposes a Causality-aware Deconfounded NER network (CDNER). By integrating multi-granularity feature extraction, causal deconfounding, and prototype learning into the CRF (Conditional Random Field) model, the network enhances the model's recognition accuracy and outputs more robust entity representations. Experimental results demonstrate that CDNER achieves performance close to the current state-of-the-art, especially excelling in complex text environments.

Paper Type: Long

Research Area: Information Extraction and Retrieval

Research Area Keywords: Information Extraction,Machine Learning for NLP,NLP Applications,

Contribution Types: Publicly available software and/or pre-trained models

Languages Studied: Chinese

Submission Number: 240