The STAB Project

Story Construction in Intelligence Analysis
Project Team: Ashok Goel, Summer Adams, Neha Sugandh and Stephan Prockow



The research issue is how might an interactive intelligence assistant abduce
a coherent story from large amounts of data, where the data is heterogenous,
constantly evolving, and often conflicting.

Goals:
1. Develop computational techniques for recognizing story plots
connecting (apparently isolated) events
2. Recognize story plots early enough to make useful
predictions about future (and past) events

Data:

• The VAST dataset synthesized by PNNL
• Includes news stories, photographs, reports, phone log, maps, voter registry
• Covers numerous events, including over 100 pertaining to illegal or unethical activities

Knowledge Representation:

• Generic activity patterns represented as hierarchically-organized

task-method structures in a Task-Method-Knowledge Language (TMKL)

• Input events and output predictions also represented in TMKL

Method:

• Inputs are instances of generic structures in stored patterns
• Patterns invoked when input event matches generic structure
• Confidence in invoked pattern increases with additional event to pattern matches
• Confidence decreases if event violates expectations

Preliminary Results:

• STAB – running system that works for illegal and unethical events in the VAST dataset
• Makes predictions both about past and future events based on invoked patterns
• First version of graphical user interface complete

Current Status:

• Plans to integrate STAB with dataforaging component for prediction verification
• Plans to implement the ability to compose story plots from primitive tasks
• Re-vising interface to STAB for enhanced interactivity

The image below represents a story pattern that has been invoked in the STAB system based on input from news stories from the VAST data set. STAB is designed to recognize story plots connecting (apparently isolated) events and to use the plots to make predictive hypotheses about future events.



Publications

This research is supported by a grant from the Department of Homeland Security's NVAC Program and is one of a number of projects from the Southeastern Regional Visualization and Analytics Center.

---

Last modified 23 April 2007 at 12:37 pm by vector