StyloLIT: Stylometry and Location Indicative Terms Based Geographic Location Estimation Using Convolutional Neural Networks

Estimation of geographic location information of users from social media portals such as twitter plays a vital role in areas such as disaster management, marketing, cyber forensics etc. At the same time increasing data privacy concerns forced the social media sites to make the sharing of geographic location as the opt-in feature, also increasing user awareness about privacy prevents the users from disclosing their location details. However, most users leave footprints unknowingly that could be used to identify their approximate location information. Since it is observed that social media users from multiple locations possess diversity in their expression of language, we propose a two level approach involving stylometry and location indicative terms to address this problem. Experimental results shows that our approach outperforms the current state of the art in predicting the geographical location of twitter users purely based on their text content.