Learning Ideological Latent space in Twitter
URL
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.
Authors
Date
2017-08-28
Department
Major/Subject
Machine Learning and Data Mining
Mcode
SCI3015
Degree programme
Master’s Programme in Computer, Communication and Information Sciences
Language
en
Pages
47 + 5
Series
Abstract
People are shifting from traditional news sources to online news at an incredibly fast rate. However, the technology behind online news consumption forces users to be confined to content that confirms with their own point of view. This has led to social phenomena like polarization of point-of-view and intolerance towards opposing views. In this thesis we study information filter bubbles from a mathematical standpoint. We use data mining techniques to learn a liberal-conservative ideology space in Twitter and presents a case study on how such a latent space can be used to tackle the filter bubble problem on social networks. We model the problem of learning liberal-conservative ideology as a constrained optimization problem. Using matrix factorization we uncover an ideological latent space for content consumption and social interaction habits of users in Twitter. We validate our model on real world Twitter dataset on three controversial topics - "Obamacare", "gun control" and "abortion". Using the proposed technique we are able to separate users by their ideology with 95% purity. Our analysis shows that there is a very high correlation (0.8 - 0.9) between the estimated ideology using machine learning and true ideology collected from various sources. Finally, we re-examine the learnt latent space, and present a case study showcasing how this ideological latent space can be used to develop exploratory and interactive interfaces that can help in diffusing the information filter bubble. Our matrix factorization based model for learning ideology latent space, along with the case studies provide a theoretically solid as well as a practical and interesting point-of-view to online polarization. Further, it provides a strong foundation and suggests several avenues for future work in multiple emerging interdisciplinary research areas, for instance, humanly interpretable and explanatory machine learning, transparent recommendations and a new field that we coin as Next Generation Social Networks.Description
Supervisor
Gionis, AristidesThesis advisor
Garimella, KiranKeywords
filter bubble, Matrix factorization, twitter, polarization, combining link and content, latent space learning