Clustering Events Based On Common Subevents

Abstract

The objective of this project is to create a methodology to find communities of similar events based on their context, which is represented by their subevents. The similarity is measured through a new metric we propose, which takes into account the similar subevents between events. The motivation behind clustering these events into larger labeled groups is to enrich the overall understanding of each individual event. The event and subevent relationships have been extracted using a weakly supervised event acquisition method and have been stored in a knowledge base. Using these pairs and the idea of hierarchical event representation, we cluster the events, which will provide insights on the similarities and differences between events in context.