OBJECTIVE:
To lay the foundation for a theoretically justifiable, computationally efficient framework to extract meaning from digital (natural language) data, to fuse this information to better exploit available intelligence, and to assess the reliability/credibility of source/evidence.
CHALLENGES:
Natural language (NL) conversion (soft data are often in the form of text).
Difficulties associated with simple probabilistic models (non-numerical models of imperfections, uncertain implications).
Reliability/credibility of source/evidence.
Non-identical scopes of evidence sources.
Imperfections in soft sources (semantics).
Extracting meaning across multiple sources.
Potentially contradictory evidence.
Computational issues.
OUR APPROACH:
Incorporates natural language processing (NLP) algorithms for extraction of richer semantics from soft data.
Employs Dempster-Shafer (DS) theory for modeling and combination of data possessing a wide variety of imperfections.
Provides a smoother transition to conventional probabilistic models.
Introduces methods for efficient processing within the NLP and mathematical modeling components.
Uses new methods for assessing evidence credibility and source reliability.