Tweet Based Geo-Location Prediction of Twitter Users using RNN

Abstract

Using a recurrent neural network architecture, we propose a model for the regional classification of the location of social media users. The model comprising RNN is used due to the fact that multilayer neural network does not take the order of the words in the tweets into account which can be a good feature to explore for predicting the location. This model takes preprocessed tweets as inputs and outperforms the current state of the art systems. A new approach has been used to train the model for this specific task, in which the word-to-vec embeddings of one model is used to train another model. This approach gives a considerable improvement in the results.