Abstract

This final report describes the work performed by the GAIA team under DARPA's AIDA program. The GAIA team was a collaboration between: (1) The University of Southern California Information Sciences Institute (USC-ISI, PI: Marjorie Freedman); (2) The University of Florida (U of F, PI: Daisy Wang); (3) Rensselaer Polytechnic Institute (RPI, PI: Heng Ji) and The University of Illinois Urbana Champagne (UIUC, PI Heng Ji); and (4) Columbia University (Columbia, PI: Shih-Fu Chang). The AIDA programs research focused on extracting and fusing multilingual, multimodal- content and synthesizing the information to identify conflicting claims. We explored diverse extraction targets in the form of knowledge elements (KEs) about events, relations, entities, sentiment, and claims. We explored knowledge fusion of both named and non-named concepts. We explored a variety of ontology targets from relatively small (derived from previous program's definitions) to enormous (with all of Wikidata as target). Software developed under this effort has been delivered to AFRL as dockized components integrated to run the full claim detection system. In addition to this report, the GAIA project resulted in numerous publications.

Date

November 1, 2023

Authors

Marjorie Freedman, Shi-fu Chang, Filip Ilievski, Heng Ji, Kathleen Mckeown, Daisy Wang