Analyzing the Impact of Sentiment and Textual Characteristics on User Engagement in YouTube Comments Across Diverse Video Genres

Understanding user engagement metrics on social media platforms is crucial for content creators, digital marketers, and algorithm optimization. This paper presents an automated, multi-stage pipeline to extract, process, and analyze YouTube comment data to determine the factors driving user engagement, specifically measured through comment “likes” and “replies.” Utilizing the YouTube Data API v3, we collected and processed comments across five distinct video categories—Discussion, Gaming, Motivation, Music, and Tutorial—yielding a final sample of N = 808 valid comments after sanitization. We applied Natural Language Processing (NLP) preprocessing and employed the VADER (Valence Aware Dictionary and sEntiment Reasoner) tool to classify comment polarity, followed by Spearman rank-order correlation analysis with Benjamini–Hochberg false-discovery-rate (FDR) correction applied across all k = 18 hypothesis tests. Our findings challenge common assumptions: comment length shows no significant correlation with likes (ρ = 0.052, p = 0.142, q = 0.279), while sentiment shows a statistically significant, albeit weak, negative correlation globally (ρ = −0.101, p = 0.004, q = 0.012). Per-video analysis further reveals that engagement dynamics are context-dependent: negative sentiment is significantly associated with higher engagement in the Tutorial video (ρ = −0.226, p = 0.007, q = 0.017), whereas the Motivation video is dominated by positive sentiment with no significant sentiment–engagement relationship (p = 0.539, q = 0.539). Replies are the strongest and most consistent predictor of likes across all five video categories (ρ = 0.190–0.384, all q < 0.05). Our pipeline provides a replicable framework for comment-level engagement analysis on YouTube.