Neural geolocation prediction in Twitter

Inferring the location of a user has been a valuable step for many applications that leverage social media, such as marketing, security monitoring and recommendation systems. Motivated by the recent success of Deep Learning techniques for many tasks such as computer vision, speech recognition, and natural language processing, we study the application of neural models to the problem of geolocation prediction and experiment with multiple techniques to analyze neural networks for geolocation inference based solely on text. Experimental results on the dataset suggest that choosing appropriate network architecture can all increase performance on this task and demonstrate a promising extension of neural network based models for geolocation prediction. Our systematic extensive study of four supervised and three unsupervised tweet representations reveal that Convolutional Neural Networks (CNNs) and fastText best encode the the textual and geoloca- tional properties of tweets respectively. fastText emerges as the best model for low resource settings, providing very little degradation with reduction in embedding size.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2017

Author dc:creator

• Srinivasan, Pramod

Contributors dc:contributor

• Zhai, ChengXiang

• Deep learning
• Information retrieval

Statement dc:rights

• Copyright 2017 Pramod Srinivasan

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/97632

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/97632