• css.php

    Visual Earth: Exploring the Content of Image Sharing Around the World

    by  • June 4, 2018 • 2017-2018 Provost Digital Innovation Grant Winners, PDIG17-18, Provost's Digital Innovation Grants

    Project Name: Visual Earth: Exploring the Content of Image Sharing Around the World

    Grantee: Agustín Indaco

    Discipline: Economics

    Funding Cycle: 2017-2018

    Project Status:

    White Paper: indaco_pdig2018

     

    About the Project

     

    Over the years, technology has continually influenced the way people communicate. Today cellphones and apps top the list of preferred communication methods. But in the last few years, advances in technologies and the proliferation of smartphones have led the way to an expansion in image-sharing as a form of communication. By analyzing 270 million image tweets sent all over the world, our latest project (Visual Earth) will show that this phenomenon is worldwide. I would like to extend this research by analyzing the content of these images shared worldwide. I plan to use deep learning methods to (i) identify the correct tags for the images, and (ii) study how the content of image-sharing differs across continents, countries, and cities. I would also develop a detailed blog post on how to implement deep learning methods in scholarly projects. This will help me track my progress and provide a valuable and open resource for other scholars who might want to use such methods in future projects.

     

    Visual Earth is the first study to analyze the growth of image-sharing around the world. Using several techniques and metrics, we measure the growth of image-sharing around worldwide, and relate this to economic, geographic, and demographic differences. The reason we found it particularly interesting to study this through Twitter is that Twitter was launched as a predominantly text-based platform. Studying image-sharing on Twitter is a way to indicate the extent to which image-sharing has become increasingly prevalent in dominant forms of communication nowadays.

     

    In analyzing the content of the images, I would like to employ a deep network method for image analysis in order to identity the categories and themes being shared on these image tweets. I am looking into the Clarifai (https://www.clarifai.com) library in R, where one can train one’s own model and choose up to 10 custom concepts. This will allow me to dig deep into understanding what type of images people are sharing on Twitter, as well as the differences in what is being shared in different parts of the world. I would include these results as an extension to the Visual Earth study in the project website. Although Clarifai, as well as similar tools I have looked at, offer a free trial version for analyzing up to 10,000 images, one needs to pay for larger operations, which I would use this grant for.

    About