Temporally Stable Multilayer Network Embeddings: A Longitudinal Study of Russian Propaganda

TL;DR

This study analyzes 2.4 million articles from RT across seven languages from 2006 to 2023, revealing inconsistent targeting strategies and thematic shifts in Russian propaganda content over time.

Contribution

It introduces a novel method for aligning multi-lingual, sparsely connected semantic networks to study longitudinal propaganda strategies.

Findings

RT's communication strategies lack coherence across languages.

Significant thematic differences exist between language versions.

Clusters align with key propaganda themes but show uneven focus shifts.

Abstract

Russian propaganda outlet RT (formerly, Russia Today) produces content in seven languages. There is ample evidence that RT's communication techniques differ for different language audiences. In this article, we offer the first comprehensive analysis of RT's multi-lingual article collection, analyzing all 2.4 million articles available on the online platform from 2006 until 06/2023. Annual semantic networks are created from the co-occurrence of the articles' tags. Within one language, we use AlignedUMAP to get stable inter-temporal embeddings. Between languages, we propose a new method to align multiple, sparsely connected networks in an intermediate representation before projecting them into the final embedding space. With respect to RT's communication strategy, our findings hint at a lack of a coherent strategy in RT's targeting of audiences in different languages, evident through differences in tag usage, clustering patterns, and uneven shifts in the prioritization of themes within language versions. Although identified clusters of tags align with the key themes in Russian propaganda, such as Ukraine, foreign affairs, Western countries, and the Middle East, we have observed significant differences in the attention given to specific issues across languages that are rather reactive to the information environment than representing a cohesive approach.