Dheeman Saha

Computer Science Ph.D. Student


Project: Followers Tell Who To

Abstract

Influencers are followed by a relatively smaller group of people on social media platforms under a common theme. Unlike the global celebrities, it is challenging to categorize influencers into general categories of fame (e.g., Politics, Religion, Entertainment, etc.) because of their overlapping and narrow reach to people interested in these categories. In this work, we focus on categorizing influencers based on their followers. We exploit the top-1K Twitter celebrities to identify the common interest among the followers of an influencer as his/her category. We annotate the top one thousand celebrities in multiple categories of popularity, language, and locations. Such categorization is essential for targeted marketing, recommending experts, etc. We define a novel FollowerSimilarity between the set of followers of an influencer and a celebrity. We propose an inverted index to calculate similarity values efficiently. We exploit the similarity score in a K-Nearest Neighbor classifier and visualize the top celebrities over a neighborhood-embedded space.

t-SNE plot
Fig.: The t-SNE plot is representing the FollowerSimilarity among the top-1K celebrities and the randomly selected micro-influencers.

To view the different cluster download the t-SNE image: [png][pdf]

The labels used to identify the clusters is available here: [Labels]

The accepted paper and Github repository: [Paper][Github]