SSDAU: Structured Semantic Data Augmentation for Joint Entity and Relation Extraction

Anonymous

Abstract: Joint entity and relation extraction (JERE) are highly susceptible to weak-generalization due to low-quality text on various natural language processing (NLP) tasks. Data augmentation is a common approach to enhance text quality. However, traditional data augmentation methods tend to compromise intrinsic textual logic, making it challenging to design effective augmentation techniques. In this paper, we propose a novel paradigm called Structured Semantic Data Augmentation (SSDAU) that that preserves the structured semantics of augmented text. SSDAU applies context-awareness to encode semantic entities in the text, then designs a Tarjan algorithm to complete the neighborhood semantic analysis of related entity, and finally completes the data augmentation by entity reconstruction of neighboring text with a structured semantic model. We compare different baselines under different JERE scenarios and evaluate the performance and efficiency of SSDAU. The results demonstrate that SSDAU can generate high-quality data, which greatly improves the performance of the JERE model and outperforms the state of the art.

Paper Type: long

Research Area: Generation

Contribution Types: Approaches to low-resource settings

Languages Studied: Data Augmentation, Structured Semantic, Neighborhood Semantic Analysis