Generating Alternatives for Interpretation and Analysis (GAIA)

Co-PI: Daisy Zhe Wang

Sponsor: DARPA

Start Date: January 4, 2018

End Date: July 7, 2022

Amount: $116,900

Abstract

We analyze and combine structured knowledge elements from event-centric knowledge graphs automatically constructed from social media sources (text, video, audio, images) to generate a set of salient alternative hypotheses for a given query. We further establish semantic consistency and coherence metrics based on probabilistic modeling, which will serve to rank the hypotheses. Alternative hypotheses generation over automatically constructed knowledge graphs generated at real time as events evolve can help sift through multi-modal and multi-language information to generate human understandable information and to help decision making.