Do Actions Speak Louder Than Words? Predicting Influence in Twitter using Language and Action Features

This work explores the connection between language, personality, and influence in a social media network. It clusters users based on two types of features: account activity features and stream content (word) features and compares the usefulness of these different types of features in categorizing users according to their influence and leadership potential in the network. Results of clustering using different sets of features are examined to answer questions about distribution of Twitter users from the influence perspective. These results are compared against distributions of personality traits obtained from previous research on personality types and established assessment tools that measure leadership aptitude and style. Experiments with different clustering algorithms are described and their performance and cluster outputs are reported.