BNU: A BALANCE-NORMALIZATION-UNCERTAINTY MODEL FOR INCREMENTAL EVENT DETECTION

Event detection is challenging in real-world application since new events continually occur and old events still exist which may result in repeated labeling for old events. Therefore, incremental event detection is essential, which requires a model to learn new events incrementally and meanwhile prevent the model performance from degrading on old events. However, existing incremental event detection methods cannot handle data imbalance problem between old and new events, and face knowledge transfer problem which cannot adequately utilize the knowledge provided by previous model and data. In this paper, we propose a Balance-Normalization- Uncertainty (BNU) model to address the above issues. Specifically, in order to mitigate the adverse effects of data imbalance, we incorporate a balanced fine-tuning stage and a cosine normalization module. Meanwhile, we consider aleatoric uncertainty to preserve previous knowledge while training for new events. Extensive experiments show that the proposed method resolves the above challenges effectively, outperforming the state-of-the-art methods by a substantial margin on ACE and TAC KBP datasets.