Overcoming the AI “Black Box” in Political Analysis: A Hybrid Approach

NLP techniques are increasingly valuable for Political Science, especially in computational text analysis, due to the growing volume of digital text data. Machine learning, particularly neural networks, has boosted NLP's capacity, often surpassing human abilities. However, neural networks' "black box" nature poses challenges for Political Science, limiting causal insights, leading to inconsistent results, and introducing bias from external datasets. To address these issues, a "hybrid" approach combines neural networks with human oversight and structured knowledge. This paper applies such a method to analyze the "Great Replacement" conspiracy theory, which has fueled societal polarization and violence. Using computational and political discourse analysis, we build an annotated corpus and apply transformer-based AI to assess polarity in related discourse. The corpus serves as a gold standard to evaluate AI performance.