Event Schema Induction

Event schema induction aims to automatically learn the typical structure and progression of events, moving beyond manually defined schemas with limited coverage. Current research focuses on leveraging large language models (LLMs) and graph-based methods, including diffusion models and prompt-based approaches, to generate and refine event schemas from text data, often incorporating incremental prompting and verification techniques. This automated schema induction promises to improve various natural language processing tasks, such as event extraction and knowledge representation, by providing more comprehensive and adaptable models of real-world events. The resulting schemas are evaluated based on their completeness, accuracy, and readability, with a focus on handling hierarchical and temporal relationships between events.