E-Locus - Institutional Repository of the University of Crete

Home Collections School/Department School of Sciences and Engineering Department of Computer Science Post-graduate theses

Post-graduate theses

Current Record: 62 of 824

[Add to Basket]

Identifier

000446946

Title

Multi-layer bipartite structural features to analyze YouTube Social Network

Alternative Title

Δομικά χαρακτηριστικά πολυεπιπέδων διμερών γράφων για ανάλυση του κοινωνικού δικτύου του YouTube

Author

Οικονομίδου, Μαρία Π.

Thesis advisor

Πρατικάκης, Πολύβιος

Reviewer

Τόλλης, Ιωάννης
Σαριδάκης, Χαράλαμπος

Abstract

This work investigates interactions on YouTube, concerning predicting missing or unseen interactions on multi-layer bipartite networks. More precisely, given a set of own interactions between YouTube users and videos, we measure how accurately we can predict comment interactions. We propose structural bipartite features, which enhance the performance of simple prediction models, to find missing or unseen links. Experimental validation of the proposed approach is carried out on multi-layer networks formed on YouTube. We have crawled an extensive dataset of YouTube videos, the channels that own them, and the authors of their comments. Using a machine learning framework, we find that we can predict future and unseen comment interactions on YouTube videos with precision 99%. We also show that to predict a day’s comment interactions it suffices to account network information generated 1 day prior. Our set-up is implemented on the MapReduce model. We propose two MapReduce algorithms, one that counts the bitruss number of an edge and one that clusters edges into blooms in a bipartite network.

Language

English

Subject

Big data

Bipartite graphs

Graph analysis

Link prediction

Διμερείς γράφοι

Μεγάλα δεδομένα

Πρόβλεψη ακμών

Issue date

2022-03-18