Please use this identifier to cite or link to this item:
http://artemis.cslab.ece.ntua.gr:8080/jspui/handle/123456789/19215
Title: | How to Go Viral: Leveraging Graph and Semantic Counterfactual Algorithms |
Authors: | Κιούρα, Ιωάννα Στάμου Γιώργος |
Keywords: | YouTube Viral videos Knowledge Graphs Semantic Counterfactuals Γράφοι Γνώσης Εξηγήσεις με αντιπάραδειγμα |
Issue Date: | 18-Jul-2024 |
Abstract: | This thesis explores the complex field of content virality and its determinant factors, specifically focusing on YouTube videos. The introductory chapters cover fundamental ideas in the essential concepts of graph theory, sentiment analysis, image captioning, semantic similarity, and counterfactual explanations, which are used to construct a comprehensive framework for understanding viral content. More specifically, graph theory basics provide the structural foundation, highlighting how graphs can represent complex relationships and dependencies present in video material. Sentiment analysis techniques are examined to understand the perception of and emotional response towards textual data. The chapter on image captioning demonstrates the integration of computer vision and natural language processing to automatically generate descriptive metadata for video thumbnails. Semantic similarity is also utilized in order to be able to compare textual data and a semantic counterfactual algorithm is issued to calculate the differences and distance between two graphs. The objective of this thesis is to identify key factors that are recurrent across viral videos and construct a framework that will provide content creators with useful advice to increase their videos' chances for virality. The proposed method involves creating a customized dataset from YouTube Trending Video Dataset, transforming video-related data into graph representations, and employing graph counterfactual algorithms to compare them with one another and identify key elements that drive a video from non-viral to viral status. Experiments are conducted, in specific video categories and in a mixed dataset as well. The most prevalent differences between non-viral and viral videos are highlighted through statistical analysis and a qualitative analysis suggests changes to non-viral example videos and explores the framework's strengths and weaknesses. Overall, this thesis provides a robust framework for understanding and enhancing viral YouTube content, combining theoretical insights with practical applications to offer valuable advice for content creators, businesses, influencers, etc aiming to maximize their reach and impact. |
URI: | http://artemis.cslab.ece.ntua.gr:8080/jspui/handle/123456789/19215 |
Appears in Collections: | Διπλωματικές Εργασίες - Theses |
Files in This Item:
File | Description | Size | Format | |
---|---|---|---|---|
Diploma_Thesis_Ioanna (2).pdf | 6.42 MB | Adobe PDF | View/Open |
Items in Artemis are protected by copyright, with all rights reserved, unless otherwise indicated.