Date of Award

5-19-2023

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Information Science

First Advisor

Nitin Agarwal

Abstract

YouTube, the leading video-sharing social media platform, is home to a vast number of videos, with over one billion hours of content consumed daily. However, with 500 hours of fresh content added to the platform every day, ensuring the quality of the content has become a daunting task, making it challenging to prevent the spread of low-quality information. The manual review of malicious content by moderators is time-consuming and highlights the need for faster analysis and processing tools. This paper proposes a novel approach that utilizes Generative Pre-Trained Transfer-3 (GPT-3), with 175 billion parameters, to extract narratives from YouTube videos. The proposed model achieved a semantic similarity of 70% with state-of-the-art frameworks designed for narrative extraction. Furthermore, the study addresses the issue of content exploration on YouTube, as viewers tend to be recommended similar videos by the platform's algorithm. To address this issue, we propose the integration of the extracted narratives into a web-based visualization tool that enables users to explore different variations of the same or combined keywords. The proposed tool aims to enhance the quality of the content and provide viewers with a more diverse range of video content. The results of our study demonstrate the potential of using GPT-3 for narrative extraction on YouTube and provide a new approach for exploring video content on the platform.

Download

Included in

Social Media Commons

Share

COinS