The Evolution of Topological Concepts in Online Social Networks

Online Social Networks (OSNs) enable individual users to connect with others, contribute to discussions and share their content. The data created in these platforms contain relationships between users and user generated content. User relationships can be represented by a social network, while user generated content takes the form of text, images and video. There are unique challenges presented by the nature of data from OSNs. Users interact with each other in complex ways at scale, which presents challenges in characterising the most important properties for new connections forming. Much of the text generated by users in OSNs is short and contains misspellings, abbreviations, acronyms and other forms of noise and domain specific idioms. This data also lacks consistent labels that can be used to train supervised learning models. Lastly, the data arrives in a stream which requires compressed representations for historic data and the ability to handle changes in the relationships present in the data over time. This thesis addresses the challenges of modelling evolving topological concepts in network and text data from OSNs by making four contributions to knowledge. Firstly, a model is developed to predict the formation of links in social networks that can discover the relative importance of network properties that characterise the network. This method is evaluated on synthetically generated network data and real-world citation networks. Secondly, document clustering methods for OSNs are developed using language embedding models to represent the textual content and unsupervised clustering algorithms to discover latent topics. These methods are evaluated on a clustering task on three real-world OSN data sets from Twitter and Reddit. Thirdly, a method to derive the hierarchical topology of topics in OSNs is developed by combining the Mapper algorithm from topological data analysis with state-of-the-art language models and dimensionality reduction techniques. This method is evaluated on a clustering task on two data sets from Twitter. Results are compared to state-of-the-art methods and an improvement is demonstrated. Finally, a stream clustering algorithm referred to as MapperStream is developed to discover and track the evolving hierarchical topology of topics in OSNs over time. This contribution combines aspects of the first three contributions with extensions to address the challenges of data streams. It is evaluated on a stream clustering task over a Twitter data stream with results exceeding the state-of-the-art. The method is applied to another Twitter data stream collected over the period of the 2022 Australian federal election campaign.

Author dc:creator

• Curiskis, Stephan A,

Statement dc:rights

• info:eu-repo/semantics/openAccess
• The author owns the copyright in this thesis including all reproduction and reuse rights for the work. The work may not be altered without the permission of the copyright owner. Attribution is essential when quoting or paraphrasing from this thesis.
• © 2022 Stephan A. Curiskis
• au.edu.uts.lib/cph

Language dc:language.iso

en_US

Handle dc:identifier.uri

http://hdl.handle.net/10453/171506

OAI identifier oai:identifier

oai:opus.lib.uts.edu.au:10453/171506