Age Inference on Twitter using SAGE and TF-IGM

Abstract

Social media is increasingly influential in day-to-day life. People are more than ever sharing, posting, liking, and following different activities on disparate social media. Deriving specific attributes of users based on their online behavior is a growing research field. In this study, a novel methodology is proposed for determining the age of Twitter users. We classify three separate age groups, namely, 18--24, 25--54, 55 >. We compute numerous linguistic features from the tweets of users, obtain significant terms extracted by the SAGE algorithms, and retrieve relevant meta-data of users by extracting information on their followed interests on Twitter using TF-IGM. The final logistic regression model obtains a macro F1-score of 78%. This way, effectively combining NLP and IR techniques for attribute inference on social media.