About
Exploring Big Visual Data
How do we explore social media’s visual data which contains billions of photographs shared by hundreds of millions of contributors? What insights can we gain from this type of massive collective visual production?
Phototrails is a research project that uses experimental media visualization techniques for exploring visual patterns, dynamics and structures of planetry-scale user-generated shared photos. Using a sample of 2.3 million Instagram photos from 13 cities around the world, we show how temporal changes in number of shared photos, their locations, and visual characteristics can uncover social, cultural and political insights about people’s activity around the world.
The project is part of the emerging research field of Cultural Analytics which uses computational methods for the analysis of massive cultural datasets and flows.
The software used to create all media visualistions in this project is avalable here: http://lab.softwarestudies.com/p/imageplot.html
All-in-One Visualizations
Typical visualization tools show data as points, lines, and bars. Our visualizations show the actual images in the collection.
We present visualizations techniques that show a large numbers of images in a single visualization and enable the exploration of both the photos’ metadata (upload dates, filters used, spatial coordinates) and the patterns created by their content.
We can scale the images to any size and organize them in any order, presenting, for example, all the images in a collection sorted by to their dates, locations, or visual characteristics.
Multi-Scale Reading
Phototrails explores the world’s photos on multiple spatial and temporal levels, moving between the planetary-scale cultural and social patterns to the micro level of and times.
As opposed to privelleging “close reading” (analysis of singular texts) or “distant reading” (analysis of large scale patterns), we use interactive visualizations to perform a multi-scale reading – moving between the global-scale cultural and social patterns and the close-ups revealing patterns of individual users.
While many visualization techniques have already been developed, paradoxically they are better at the macro levels than the micro views. Our multi-scale “data-vision instruments” overcome this problem, allowing exploration of both singular images and patterns across tens of thousands of images.