Developing Pattern and Anomaly Detection Methods in Influence Campaigns
Name
mitchell-williamm-meng-eecs-2023-thesis.pdf
Description
Thesis PDF
Size
3.23 MB
Format
Adobe PDF
Checksum (MD5)
cd3bdf64ddafb6c3c03d9f3781db43a7
Author(s)
Mitchell, William B.
Advisor(s)
Harnasch, Raul
Ross, Dennis
Date Issued
September 2023
Publisher
Massachusetts Institute of Technology
Abstract
Influence operations are a prominent psychological component of modern warfare. Recent historic events including the 2016 US election, the 2021 Myanmar Coup, and the Russian invasion of Ukraine in early 2022 have catapulted Department of Defense
(DoD) interest in modeling and predicting outcomes of military and political events, particularly in regions of strategic interest to the US. The MIT Lincoln Laboratory, Group 52, under contract with USTRANSCOM, has developed the Global Influence Model (GIM) to evaluate the information landscape at scale. This project seeks to expand on the previous work of Group 52 on GIM, incorporating pattern and anomaly detection methods. Several statistical and machine learning methods were applied to a data set of approximately 30,000 news articles from a 2-year period between August 2019 and August 2021. Statistical methods included moving average models and Singular Spectrum Analysis (SSA). Machine learning techniques included the use of an autoencoder and an LSTM neural network. These methods provide different ways to visualize and characterize the data. Together, the approaches offer a holistic picture of events in specific countries over a time period of interest. The figures generated by these techniques may be a useful tool for a military intelligence analyst. These products allow for the rapid visualization of large news article data sets that can help model influence campaigns as they unfold.
MIT Department
Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science
Terms of Use
In Copyright - Educational Use Permitted
Copyright retained by author(s)
Persistent DSpace Link
