A Review of Continual Relation Extraction
Chen Lixi, Li Jianping, Luo Jialan, Jia Chen, Cui Changrun · 2023
In the digital era, the abundance of networked data has sparked interest in Information Extraction (IE), specifically Relation Extraction (RE). RE, uncovering connections between subjects and objects, faces limitations in traditional models confined to fixed relation sets. These models struggle to adapt to evolving entity relationships in real-world scenarios. Continual Learning (CL) presents a pivotal moment for RE, offering adaptability and flexibility. CL enables models to adapt to changing data distributions, supporting incremental learning for efficient adaptation to new tasks. Researchers explore this integration, known as Continual Relation Extraction (CRE), aiming to address these limitations. This paper surveys CRE methods and explores their potential applications.