Detecting and Aligning Narrative Variants for Countering Disinformation
Abstract
In this project, we will develop new technology for detecting and analyzing disinformation in information streams. The basic premise"" of the work is that an adversary will be trying to push its own ~narrative~ of events to serve its strategic goals, and to do this"" it will introduce specific narrative variants in the form of alternative events, facts, or statements. These narrative variants may"" range from a single misleading statement carefully calibrated to sow doubt, to a set of statements that presents an alternative, mo""re favorable interpretation of accepted events, to a completely alternative set of facts that contradict reality. Building on prior,"" DARPA-funded work on the Analogical Story Merging (ASM) algorithm, we will develop new approaches to detect and analyze these varia""nts. First, we will develop methods to classify documents in information streams as story variants: does a specific document contain"" a variant of the story which we aretracking? Second, we will develop techniques for aligning a story variant with related variants"",including potentially the ~canonical~ or ~target~ narrative that has been separately identified as ground truth. This alignment pr""ocess will involve identifying the major narrative breakpoints as well as detailed event differences, using an augmentation of the e""xisting ASM approach. Third, we will test our approach on several sets of data, which we will annotate for ground truthalignment. W"e will begin with existing sets of story variants which have already been collectedand are well understood. We will also undertake" major data collection to collect story variants forgeopolitical events over the past two decades, including the Crimean crisis and"" the Estonian cyberconflict, which we will also annotate and use to test our approach. Finally, we will set the stage for future wo""rk by developing and applying a classification scheme of communicative strategies to the parts of the variants, and begin building c""lassifiers for assigning disinformation strategies to specific communicative acts: for example, is a statement intended to dismiss k""ey narrators of the target narrative, so as to undermine their authority? Is it intended to distort the underlying evidenceof the t""arget narrative, so as to diminish its power? Is it intended to distract from the target narrative by changing the topic? Or perhaps"" is it intended to dismay potential actors by introducing previously unimagined, negative consequences?
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 29, 2017
- Source ID
- N000141712983
Entities
People
- Mark Finlayson
Organizations
- Florida International University
- Office of Naval Research
- United States Navy