M. Shahriar Hossain, Patrick Butler, Arnold Boediardjo, Naren Ramakrishnan

Abstract

Intelligence analysts grapple with many challenges, chief among them is the need for software support in storytelling, i.e., automatically 'connecting the dots' between disparate entities (e.g., people, organizations) in an effort to form hypotheses and suggest non-obvious relationships. We present a system to automatically construct stories in entity networks that can help form directed chains of relationships, with support for co-referencing, evidence marshaling, and imposing syntactic constraints on the story generation process. A novel optimization technique based on concept lattice mining enables us to rapidly construct stories on massive datasets. Using several public domain datasets, we illustrate how our approach overcomes many limitations of current systems and enables the analyst to efficiently narrow down to hypotheses of interest and reason about alternative explanations.

People

patrick-updated

Patrick Butler


Ramakrishnan-updated

Naren Ramakrishnan


Publication Details

Date of publication:
August 12, 2012
Conference:
ACM SIGKDD International Conference on Knowledge Discovery and Data Mining