Breaking the Cycle of Radicalization

Project Abstract

In the digital era, information control and manipulation have found new grounds in online platforms, where influential actors strategically shape public discourse and civic engagement. Our project revolved around the critical endeavor of detecting hidden influence in dynamic networks, with a unique focus on leveraging temporal signals alone, rather than message content, to reveal coordinated activities. By analyzing the flow and timing of messages, we offered a novel perspective on understanding the dynamics of hidden influence.

Our project’s unifying thread is the emphasis on temporal analysis alone, focusing solely on the dynamics of interactions without inspecting the message contents. This approach not only augments our understanding of hidden influence in digital networks but also holds significant implications for combating propaganda campaigns and safeguarding fact-based reasoning and trust in institutions.

Project Results

This project comprises three distinct but interconnected studies.

Disambiguation of Accounts:

The first study involved the disambiguation of accounts linked to the Chinese Communist Party from organic accounts. Drawing from the influence model, a generative model that describes the interactions between networked Markov chains, we successfully distinguished orchestrated activity from organic behavior within a time-evolving network. We also implemented and open-sourced a new, more performant, version of the influence model for future researchers. This work was awarded the Best Paper award at the International Conference on Social Computing, Behavioral-Cultural Modeling and Prediction and Behavior Representation in Modeling and Simulation (SBP-BRiMS).

Identification of Accounts:

We delved into the identification of accounts associated with Russia’s Internet Research Agency, another source of global influence campaigns. We highlighted coordinated activities by analyzing temporal correlations in the sequential decision processes of individual social media accounts. This second study further solidified the effectiveness of our approach in uncovering hidden influence.

Diplomatic Impact:

Building upon the insights gained from the influence model, the third study explored a unique dimension of influence—how Russian and Chinese diplomats impact one another’s online activity. By harnessing node embeddings of topic graphs, we discerned the intricate relationships within diplomatic interactions.

A visual representation of the node embeddings of diplomat accounts’ discourse using t-SNE, a dimensionality reduction algorithm, in a two-dimensional mapping space. The average Euclidean distances of the vector representations of the two sets of accounts show a threefold reduction over the three-month period of analysis, revealing increasing alignment in diplomatic rhetoric between the two countries. (Erhardt, K. and Pentland, A., 2023).