Vote Prediction Models for Signed Social Networks
Loading...
Journal Title
Journal ISSN
Volume Title
Perustieteiden korkeakoulu |
Master's thesis
Unless otherwise stated, all rights belong to the author. You may download, display and print this publication for Your own personal use. Commercial use is prohibited.
Author
Date
2020-06-16
Department
Major/Subject
Machine Learning, Data Science and Artificial Intelligence
Mcode
SCI3044
Degree programme
Master’s Programme in Computer, Communication and Information Sciences
Language
en
Pages
81
Series
Abstract
Voting is an integral part of the decision-making mechanism in many communities. Voting decides which bills become laws in parliament or users become administrators on Wikipedia. Understanding a voter's behaviour and being able to predict how they will vote can help in selecting better and more successful policies or candidates. As votes tend to be for or against a particular agenda, they can be intuitively represented by positive or negative links respectively in a signed network. These signed networks allow us to view voting through the lens of graph theory and network analysis. Predicting a vote translates into predicting the sign of a link in the network. The task of sign prediction in signed networks is well studied and many approaches utilize social theories of balance and status in a network. However, most conventional methods are generic and disregard the iterative nature of voting in communities. Therefore this thesis proposes two new approaches for solving the task of vote prediction in signed networks. The first is a graph combination method that gathers features from multiple auxiliary graphs as well as encoding balance and status theories using triads. Then, it becomes a supervised machine learning problem which can be solved using any general linear model. Second, we propose a novel iterative method to learn relationships between users to predict votes. We quantify a network's adherence to status theory using the concept of agony and hierarchy in directed networks. Analogously, we use the spectral decomposition of the network to measure its balance. These measures are then used to predict the votes that comply the most with the social theories. We implement our approaches to predict votes in the elections of administrators in Wikipedia. Our experiments and results on the Wiki-RfA dataset show that the iterative models perform much better than the graph combination model. We analyse the impact of the voting order on the performance of these models. Furthermore, we find that balance theory represents votes in Wikipedia elections better than status theory.Description
Supervisor
Gionis, AristidesThesis advisor
Ordozgoiti, BrunoKeywords
signed networks, balance theory, status theory, graphs, voting models
