Home // AIMEDIA 2025, The First International Conference on AI-based Media Innovation // View article

From Metadata to Meaning: GPT-4 Reveals Bias Trends in YouTube

Authors:
Nitin Agarwal

Keywords: YouTube recommendation system; artificial intelligence (AI); GPT-4; sentiment analysis; emotion analysis; toxicity analysis; bias; narrative analysis; recommender systems; social media algorithms; human-centered AI component; Open AI Whisper model.

Abstract:
YouTube’s recommendation system significantly shapes user experiences but has raised concerns over potential bias and the formation of filter bubbles. Traditional studies have primarily relied on metadata, such as video titles, which often fail to capture the full context or nuance of video content. This study harnesses recent advancements in Artificial Intelligence (AI)—specifically the capabilities of Generative Pre-trained Transformer 4 (GPT-4)—to conduct a deep comparative analysis of sentiment, emotion, and toxicity across multiple layers of YouTube video content. By leveraging AI to extract and interpret narrative elements beyond superficial metadata, the research uncovers key patterns: a shift from neutral to positive sentiment and emotion (especially joy) with increased content depth, a consistent decrease in anger, and divergent toxicity trends—rising in titles but decreasing in deeper narrative analysis. These findings underscore AI’s transformative role in enhancing content understanding and addressing long-standing challenges in recommendation system bias.

Pages: 83 to 88

Copyright: Copyright (c) IARIA, 2025

Publication date: July 6, 2025

Published in: conference

ISBN: 978-1-68558-330-9

Location: Venice, Italy

Dates: from July 6, 2025 to July 10, 2025