UCLA

Influencer marketing, a word-of-mouth marketing strategy that leverages prominent individuals on social media, has been gaining great traction in recent years. As the number of influencers has explosively increased with the rapid growth of the influencer marketing industry, several issues in evaluating and discovering valuable influencers have been raised, including the influencer fraud problem. These issues arise from utilizing solely quantitative metrics such as the number of followers to identify influencers on social media. Although these quantitative indicators can measure the influence of social media users to some extent, it is challenging to evaluate the quality of influencers with such metrics.

In this dissertation, we propose several graph learning frameworks that incorporate various input sources from social media including text, images, and graphs, to evaluate influencers in the first place. To that end, we build and analyze social networks of influencers based on the posting, interacting, and advertising behaviors of influencers to find valuable insights by understanding the social relations of influencers. The frameworks presented in this dissertation take the influencer social networks with decent features from multi-modal inputs and then assess qualities of influencers, including transparency, loyalty, authenticity, and efficiency of influencers. As a result, the proposed methodologies not only address pressing issues in the influencer marketing industry but also advance multi-modal graph learning-based applications.