A Comparative Evaluation of URL-Sharing, Content Similarity, and Temporal Synchronicity Signals for Detecting Coordinated Inauthentic Behavior in Multilingual Political Discourse

Abstract

Coordinated inauthentic behavior (CIB) on social media platforms poses a persistent threat to the integrity of political discourse. While multiple signal types have been proposed to detect such coordinated manipulation, no prior study has systematically compared their detection efficacy across languages. This paper presents a comparative evaluation of three widely adopted coordination signals — URL co-sharing, content cosine similarity, and temporal posting synchronicity — applied to English and Russian political CIB datasets derived from the Twitter Information Operations Archive and the Clemson Internet Research Agency corpus. For each signal type, we construct a coordination network, apply Leiden community detection, and measure precision, recall, and F1 against ground-truth information operation labels. Our results indicate that URL co-sharing achieves the highest F1 in both languages (0.802 in English, 0.787 in Russian), while content similarity suffers a 16.0% F1 degradation in Russian attributable to morphological complexity. Temporal synchronicity yields the most language-stable performance with only a 2.3% cross-language gap. A late-fusion combination of all three signals reaches an F1 of 0.891 in English and 0.824 in Russian. Parameter sensitivity analysis reveals that optimal time-window configurations differ across languages and signal types. These findings offer practical guidance for platform governance teams selecting detection indicators for multilingual CIB monitoring.