From Algorithms to Actors: Detecting Foreign Interference Through Entity Extraction and LLM-Based Methods

This study leverages advanced entity extraction algorithms and large language models (LLMs) to detect Foreign Information Manipulation and Interference (FIMI) within global datasets, focusing on the ACLED and Google Trends platforms. By integrating computational linguistics with political science frameworks, the research addresses a critical gap in automated detection of disinformation and influence operations. Employing LLMs fine-tuned for multilingual and domain-specific tasks, we develop a pipeline to identify and classify FIMI-related entities, events, and narratives. First, the ACLED database is analyzed to extract geopolitical event data, targeting instances linked to disinformation or manipulation tactics. Concurrently, Google Trends data is mined to detect anomalous search behaviors indicative of coordinated influence campaigns. The pipeline incorporates Named Entity Recognition (NER) and topic modeling to flag potential FIMI instances based on predefined taxonomies, validated through cross-referenced human coding.