Mining fashion influence on social media

Social media has revolutionized the marketing strategies of the fashion industry, enabling companies to engage with a wider and more diverse audience through platforms such as Twitter and Instagram. This transformation is primarily driven by fashion influencers, accounts on social media that shape fashion trends and impact consumers' purchasing decisions. In response, fashion brands, retailers, and media companies actively seek to engage influencers aligned with their brand's goals. This dissertation explores how fashion influence can be mined on social media and better harnessed by the fashion industry. We introduce a framework for identifying top fashion influencers on Twitter (FITNet) and develop an interactive visualization tool (FITViz) to discover how fashion influence is transmitted through interactions among these accounts and the larger fashion subgraph. FITNet is the first large-scale dataset of fashion influencers on social media. It categorizes each of its 10k fashion accounts (e.g., individuals, brands, retailers, and media); and exposes the following relationships between them, as well as their retweet and mention interactions between January 1st, 2018, and February 1st, 2019. We demonstrate how FITNet can be used to answer questions about fashion influencers, including how influential they are, who they influence, who they are influenced by, and how influence is geographically distributed. To facilitate additional insights discovery over the the FITNet dataset, we developed FITViz, an interactive visualization tool that allows users to quickly curate smaller networks based on a set of seed accounts or specific fashion-related topics and study content creation and network effects within those contexts. Finally, we conducted a user study with FITViz to validate its current utility and discover new use cases. The user study explores the strengths and limitations of different facets of the tool by soliciting feedback on its capabilities, user interface, and user experience. From these findings, we distill a set of design principles for exploratory visualization tools over fashion influence data. These contributions set the stage for promising new data-driven fashion applications that combine social media data with large language and generative AI models.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2023

Author dc:creator

• Han, Jinda

Contributors dc:contributor

• Kumar, Ranjitha
• Karahalios, Karrie
• Sundaram, Hari
• Zhao, Li

• Fashion
• Influencers
• Influence Mining
• Social Networks
• Twitter
• Interactive Visualization

Statement dc:rights

• Copyright 2023 Jinda Han

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/122170

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F122170