Title: Navigating the Fine Line: A Critical Look at Conventional and Dark Humor in YouTube Shorts

Abstract

The rise of video platforms like YouTube has fundamentally altered user interaction with media, shifting focus toward concise, high-impact formats such as YouTube Shorts. Within this digital landscape, a contentious "gray area" exists: content that complies with platform guidelines yet may inadvertently cause distress to specific viewers. To investigate this phenomenon, we present TwistedHumor, a comprehensive dataset comprising 1,211 YouTube Shorts and 33,041 associated comments. Each entry has been manually annotated to capture humor presence, classification, potential harm, subject matter, rhetorical techniques, and stand-up comedy context.

Beyond data collection, this study offers a multi-perspective examination of how humor and potential harm intersect in short-form social media. By employing LLooM-based concept induction on video descriptions, we reveal that dark humor is not a monolithic category; instead, it frequently clusters around themes of identity expression, coping mechanisms, awkwardness, and social critique. Our analysis of audience engagement via linked comments indicates a divergence in reception: while conventional humor tends to elicit positive sentiment, dark humor provokes a more polarized response, characterized by mixed, neutral, and occasionally toxic reactions. Additionally, our evaluation of large language models against human annotations demonstrates superior performance in identifying stand-up comedy compared to shorter joke formats. Collectively, these findings establish TwistedHumor as both a new benchmark and an empirical exploration of the nuanced boundary between humor and harm, underscoring the necessity for context-sensitive moderation strategies and more resilient multimodal assessment tools.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC