Seeing The Spread of (Mis)information on Twitter
Rumors tend to run rampant on twitter and the Guardian just published a visualization to illustrate how these rumors evolve and are corrected in the context of the London riots. The stories featured are framed as either myths or truths and each section takes you through an individual timline of that event.
Designed by the interactive team at the Guardian, led by Alastair Dant, this visualization is functionally advanced and richly detailed in the data it presents. The tools uses a custom physics engine for the twitter circles as well as custom renderers depending on the browser. Read more about the behind the scenes in this post on the Guardian data blog.
It took four and a half people about three weeks to build (design/development) w data analysis by academic partners – @ajdant
In essence, the tool graphs tweets per hour across the top in a navigational timeline. In different places along the timeline key events are highlighted and explained in the left sidebar. The team then plotted influential tweets and retweets of related stories in an abstract space below, color coding them by the sentiment of the tweet. With much more subtlety, the ‘originating’ tweet is highlighted in blue to match the vertical bar that appears in the timeline.
Users can explore the events in more detail by hovering over individual bubbles to read the contents of a tweet. Doing so replaces the key along the bottom with the tweet while clicking keeps a tweet visible as you move about the piece.
The general interface of this tool is functional and makes sense but I think it breaks a lot of the conventions set by similar visualizations from the same team like “How Twitter tracked the News of the World scandal.“
The play button seems like an afterthought compared to the Murdoch visualization and the previous and next buttons could be made more prominent. I think the visual hierarchy is what bothers me the most. Once I click on a particular story its title and ‘authenticity’ are left aligned, though still near the top. The text of the original tweet is also quite large compared to the contextual information and labels on the visualization.
The timeline too seems more thought out in the earlier twitter visualization. The London riots timeline has 4 different types of vertical lines on it, all with a fair amount of contrast and the labels for the dates are awkwardly placed. I do like that there is a stepper in this visualization that snaps to different points on the timeline but its tricky at different points to figure out which blue line the information below is referring to. It may be nit-picky but the rendering of the blue lines is inconsistent and blurry at times, effecting readability.
The circles themselves, the crux of the visualization, give you a sense of the volume of tweets as well as a loose grouping but you have to do some digging to determining where reports or events originated. The tweet bubbles are scaled by their influence (twitter followers) and generally get smaller over time which makes sense to me. However it seems the originating tweet follows the same rules and isn’t given any extra importance in terms of scale for being the first of its kind. The blue circle around the originating tweet isn’t explained in the key and it takes some decoding of the tweet’s text to see what it means. It’s also difficult to track changes over time. There’s no quick way to see which tweets are newer and which are older. This could be done with opacity or perhaps their horizontal position in space.
It’s easy to draw parallels to a different visualization of sharing that primarily utilizes circles – Google+ Ripples. Obviously the Guardian can’t rip off the interface or functionality but there are useful concepts to reference. I think Google+ Ripples does a great job of showing influence, volume, and directionality of information over time. That’s no easy feat and I think the Guardian piece does most of this but doesn’t effectively communicate the directional flow of information. Grouping the tweets together partially accomplishes this but you can tell tweets in different groups have very similar structures but you can’t determine which came first without reading.
Admittedly twitter presents a much greater challenge when visualizing the flow of information in comparison a network like Google+. Twitter users can change text, shorten links, retweet in different ways, paraphrase and do any number of things to cloud the path of sharing. Not to mention the order of events matters a bit more. This difficulty is why I have yet to see a visualization of conversations on twitter as effective as Google+ Ripples despite the dozens of them that exist. I think twitter suggests a different visual structure than a Google+ post. The London Riots is an impressive visualization of a small subset of twitter data for a complex issue. I don’t think they hit the nail on the head for visualizing twitter data, nor were they trying to. Despite all that exist I think there’s room for one very very excellent visualization tool for twitter that shows conversations and is useful for analytics.
Who wants to start working on it?
Nice analysis of the visualization.
The scaling of the tweets over time seemed strange to me. Representing both influence and time by scale creates some confusion.
They also had the color key at the bottom which shows older tweets being darker. I found that really hard to pick out of all the circles. I actually thought all the colors in general to be less than intuitive.
Any idea of what the outside circle represents? That was really distracting for me. My initial guess is that it represents time away from the peak, but I wasn’t sure.
But looking beyond the details – I found the visualization did a pretty bad job of telling the story. If you hit play the “How the rumor unfolded” text goes by too quickly to actually read. They also fail to relate the left side to the main section.
Looking forward to more posts.
-Joe
P.S. I definitely feel the same way about the blue line!
I kind of like the fuzzy relationship between time and influence. I think it holds true. Older tweets are less influential to some degree.
The design mentioned the color darkening was a last minute addition and will be corrected in an update. It’s definitely not intuitive. Lighter would make more sense.
I couldn’t tell what the larger circle around all the tweets meant. It seemed to react strangely to the physics system. It might be total volume or something like that. Hard to tell and a good question.
I think the piece overall does tell the story well. I don’t think the default ‘play’ functionality works at all. You almost have to use the stepper or drag the playhead.
Thanks for your thoughtful comments Joe
Hmmm. Ya, I was so excited about seeing it play over time I jumped right to hitting the play button. Definitely easier to see the story if you slowly go through the timeline. Thanks for the tip.
Pingback: Expodomain.com » Revealing How Unsubstantiated Rumors Spread via Twitter
Nice post. Would be good to know a bit more about how they made it technically. You got any information on that?
We made a video accompanied by Brian Eno music using this interactive, shows the dynamics quite well and also how important it is to have questioning statements in the initial seeding process in order for rumor to exist longer. Here is the link: http://noduslabs.com/posts/the-guardian-riot-rumors-visualization/