Document-Level Event Extraction Based on Multi-instance Learning
Abstract: Current research focuses on utilizing intradocument information for event extraction, but it has limitations in capturing the complexity and diversity of events because it overlooks the relationships between documents. Additionally, current instance learning methods for event extraction primarily focus on similaritybased
instance retrieval, failing to emphasize comprehensive model learning, and a single measure of similarity cannot fully reflect the semantics of a document. To address these issues, this paper proposes an event extraction model based on multi-instance learning, exploring the connections between documents through event types and event arguments. We designed multiple instance selection strategies and construction methods to enable the model to achieve a more thorough understanding of events. Furthermore, we implemented a two-stage training approach to optimize the model’s ability to learn from instances obtained through different instances. Experiments conducted on the RAMS and WIKIEVENTS datasets demonstrate that our method surpasses the current state-of-the-art models in terms of F1 scores, validating its effectiveness and superiority.
Paper Type: Long
Research Area: Information Extraction
Research Area Keywords: event extraction
Contribution Types: Model analysis & interpretability
Languages Studied: English
Submission Number: 2828
Loading