What Propels Celebrity Follower Counts? Language Use or Social Connectivity

Follower count is a factor that quantifies the popularity of celebrities. It is a reflection of their power, prestige and overall social reach. In this paper we investigate whether the social connectivity or the language choice is more correlated to the future follower count of a celebrity. We collect data about tweets, retweets and mentions of 471 Indian celebrities with verified Twitter accounts. We build two novel networks to approximate social connectivity of the celebrities. We study various structural properties of these two networks and observe their correlations with future follower counts. In parallel, we analyze the linguistic structure of the tweets (LIWC features, syntax and sentiment features and style and readability features) and observe the correlations of each of these with the future follower count of a celebrity. As a final step we use there features to classify a celebrity in a specific bucket of future follower count (HIGH, MID or LOW). We observe that the network features alone achieve an accuracy of 0.52 while the linguistic features alone achieve an accuracy of 0.69 grossly outperforming the network features. The network and linguistic features in conjunction produce an accuracy of 0.76. We also discuss some final insights that we obtain from further data analysis celebrities with larger follower counts post tweets that have (i) more words from friend and family LIWC categories, (ii) more positive sentiment laden words, (iii) have better language constructs and are (iv) more readable.