# On Predicting Geolocation of Tweets using Convolutional Neural Networks

## TL;DR

This paper introduces a convolutional neural network model that combines tweet text and profile metadata to accurately predict the geographic location of Twitter users, significantly outperforming baseline methods.

## Contribution

The paper presents a novel CNN-based approach that integrates multiple data sources for improved geolocation prediction of tweets.

## Key findings

- Achieved 52.8% accuracy at city level
- Achieved 92.1% accuracy at country level
- Outperformed baseline geolocation methods

## Abstract

In many Twitter studies, it is important to know where a tweet came from in order to use the tweet content to study regional user behavior. However, researchers using Twitter to understand user behavior often lack sufficient geo-tagged data. Given the huge volume of Twitter data there is a need for accurate automated geolocating solutions. Herein, we present a new method to predict a Twitter user's location based on the information in a single tweet. We integrate text and user profile meta-data into a single model using a convolutional neural network. Our experiments demonstrate that our neural model substantially outperforms baseline methods, achieving 52.8% accuracy and 92.1% accuracy on city-level and country-level prediction respectively.