Continual Few-shot Event Detection via Hierarchical Augmentation Networks

TL;DR

This paper introduces continual few-shot event detection (CFED), a more commonly encountered scenario when a substantial number of labeled samples are not accessible, and proposes a memory-based framework: Hierarchical Augmentation Network (HANet).

Abstract

Traditional continual event detection relies on abundant labeled data for training, which is often impractical to obtain in real-world applications. In this paper, we introduce continual few-shot event detection (CFED), a more commonly encountered scenario when a substantial number of labeled samples are not accessible. The CFED task is challenging as it involves memorizing previous event types and learning new event types with few-shot samples. To mitigate these challenges, we propose a memory-based framework: Hierarchical Augmentation Networks (HANet). To memorize previous event types with limited memory, we incorporate prototypical augmentation into the memory set. For the issue of learning new event types in few-shot scenarios, we propose a contrastive augmentation module for token representations. Despite comparing with previous state-of-the-art methods, we also conduct comparisons with ChatGPT. Experiment results demonstrate that our method significantly outperforms all of these methods in multiple continual few-shot event detection tasks.

Figures (7)

• Figure 1: Memory-based framework for continual few-shot event detection. It preserves previous knowledge by maintaining a memory set "M." and transferring knowledge from previous models.
• Figure 2: Our system consists of a general event detector, prototypical augmentation, and contrastive augmentation. When learning new tasks with an event detector, the model replays prior knowledge from the augmented feature. Then, contrastive augmentation maximizes the acquisition of knowledge from few-shot samples.
• Figure 3: ${F1}_{micro}$ performance of every sub-task on 2-way MAVEN and 4-way ACE.
• Figure 4: Embedding space visualization via t-SNE on original and prototypical augmented feature in task $T_2$. Points within the same color indicate identical event types. As we can see, after prototypical augmentation, the intra-class distances become closer for each type. Besides, some hard samples (pointed in the squared region) initially proximate to the centers of other classes in the original space become easier to classify after prototypical augmentation, showcasing the effectiveness of prototypical augmentation.
• Figure 5: ${F1}_{micro}$ performance of each sub-task in Larger MAVEN benchmark.