Is media sentiment associated with future conflict events?
- Anne Jamison,
- Kristian Hoelscher,
- Jason Miklian,
- Witold Henisz,
- Brian Ganson
Abstract
Is media sentiment a leading indicator of conflict events? Analysis
through machine learning and natural language processing techniques
allows us to gather and process sentiment data at unprecedented scale,
depth, and accuracy. By measuring the emotional intensity and direction
of text in news reports, this baseline study shows how media sentiment
analysis can deliver new value for peace and conflict research. Using
GDELT's global sample of more than five billion media articles, we test
the relationship between sentiment data and conflict events. To achieve
both spatial and temporal precision we utilize the PRIO-GRID data
structure at daily and monthly intervals. We find that more conflictual
sentiment is significantly associated with spatially and temporally
proximate future conflict events as measured by the ACLED, SCAD and
UCDP-GED datasets. We suggest that conflict sentiment can help us
analyze conflict escalation processes more precisely.