A Pipeline for the Analysis of User Interactions in YouTube Comments: A Hybridization of LLMs and Rule-Based Methods

Abstract

This study presents a novel approach to analyze user interactions on YouTube, addressing the platform's API limitations in capturing comprehensive conversation chains. By combining Large Language Models (LLMs) and rule-based methods, we developed a pipeline to reconstruct comment threads and analyze user stances on controversial topics. We applied this approach to examine immigration debates across 27,000 comments from YouTube channels with varying political leanings. Using the E-I entropy index, we quantified cross-stance interactions. Our findings reveal that “Contra-Immigration” users dominate discussions and exhibit strong in-group communication tendencies, particularly on right-leaning channels. In contrast, “Pro-Immigration” users show a higher propensity to engage with opposing viewpoints. This research provides insights into the complex dynamics of online political discourse and potential polarization on YouTube, while demonstrating the efficacy of our hybrid approach in overcoming technical challenges in social media analysis. The study lays the groundwork for future investigations into the relationships between channel political bias, audience diversity, and user engagement patterns on controversial topics.