The measurement problem
Diffusion models describe that information spreads; influence measurement asks
which actors and relationships drove that spread. Answering that requires methods
that can attribute directional influence between individuals and communities, rather
than only describing aggregate spread curves.
Transfer entropy
Transfer entropy is an information-theoretic measure of directed influence between
time series: it quantifies how much knowing one actor’s past behavior reduces
uncertainty about another actor’s future behavior, beyond what that actor’s own
history already explains. Applied to social media activity, it lets researchers infer
directional influence pathways without assuming a fixed network structure in advance.
Pathways within and between communities
The work was carried out in the DARPA MIPs program as a cross-institutional
collaboration between UCF, North Carolina State University, and MIT. Chathura
developed methodology and an interactive dashboard that use transfer-entropy-based
methods to identify inter- and intra-community influence pathways in
information-diffusion environments.
Influence-network construction
To support this analysis at scale, Chathura developed the Influence Network
Generator (ING), an open-source Python package for constructing influence networks
from online social media data.
Interactive analysis
The Transfer Entropy Interactive Dashboard brings these methods into one
analytical system. Chathura conceived its methodology and led algorithm design,
system architecture, and visualization; it was presented as the official final
deliverable to the DARPA-MIPS program.
My contribution
Chathura conceived the dashboard’s methodology and led its design and architecture,
and was sole developer of ING. He is an author on the transfer-entropy publications
listed below.
The behavioral premise behind this work — that influence is not uniform across
interaction types — was first developed for the Multi-Action Cascade Model in the
DARPA SocialSim work. The broader line of diffusion and influence research is
described under Information Diffusion & Influence.