Continual Extraction of Semantic Relations using Augmented Prototypes with Energy-based Model Alignment

Continual relation extraction (CRE) is a critical task in natural language processing thataims to learn new relation types incrementally while preserving knowledge of previously learnedrelations. However, existing CRE models often struggle with catastrophic forgetting and inefficientutilization of memory. In this paper, we propose a CRE model that leverages class-specific prototypesand energy-based latent alignment to address these challenges. Our approach stores relation prototypes instead of real data points, enriching them with Gaussian noise during training. We incorporatecontrastive learning to obtain effective representations for memory prototype data and introduce anEnergy-based Latent feature space Alignment (ELI) module to mitigate representational shift acrosstasks. We evaluate our model on two benchmark datasets: FewRel, a balanced few-shot relationclassification dataset, and TACRED, a large-scale imbalanced relation extraction dataset. Extensiveexperiments demonstrate that our proposed method consistently outperforms state-of-the-art CREmodels across multiple tasks, with improvements of up to 4% over existing methods. This consistentsuperior performance highlights our model effectiveness in addressing the challenges of continual relation extraction, particularly in maintaining performance across a sequence of tasks while mitigatingcatastrophic forgetting.