Nicolas Belmonte
SND.org caught up with Nicolas Belmonte of Twitter about the visualization he worked on the for the State of the Union, which tracked tweets across the U.S.
Tell me about how this visualization came to be. Can we expect to see more visualizations coming out of other major news events?
Since I joined in April 2012 we released interactive visualizations around sports, T.V., politics, news and more. For example there’s the Philippines visualization, the Political Engagement Map done during the elections, the 50 year anniversary Doctor Who visualization, the Champions League visualization, etc. We change our goals depending on the event, some visualizations tend to be playful, others more insightful, in others we focus more on storytelling.
This visualization was a cross-functional team effort between the Government, Media, Comms and Visual Insights team. You’ll definitely see more of these in the future. We generally publish them from the @TwitterData account and our homepage.
How did you end up with the visualization you used? Why the streamgraph?
The streamgraph at the top of the page serves as a table of contents of what topics have been discussed during the SOTU Address. It’s a good sort of “TL;DR” approach to the SOTU transcript. You can then dive into the content by clicking on the the part of the streamgraph that interests you most. We wanted some sort of timeline and actually we tried a few different techniques. First just a classic line chart, but we realized that the added values for all the topics also was important. So we tried an area chart. The area chart showed data in a very similar way than the streamgraph but in this case it didn’t have the aesthetic value we were looking for. We have successfully used the streamgraph in other projects, so we tried it and it looked great.
What challenges did you have in building the visualization? What was the turnaround time from the end of the STOU to when you published?
The main challenge was on visualization design. There were a few different prototypes before the final version. In particular figuring out what data should we mine for and what sort of analysis would be interesting to do in that data was challenging. We agreed that the Government team would send us a list of categorized keywords. We used SOTU’s 2013 dataset and mined the data there to get topics and a timeline. We annotated the time in a few paragraphs of the SOTU 2013 address and tested the visualization with that data. After a few iterations we found that the visualization worked for us. A few hours after the 2014’s SOTU address we got the data and replaced the datasets. Of course a few final minute touches were made and we released the visualization at around 2am PT the next day.
What are you most happy with in terms of how the visualization turned out?
I’m really happy that the visualization focuses on the data itself and the design was mostly about removing clutter and leaving a clear path between the user and the data s/he’s interacting with. I love the emergence of patterns in topics and how the streamgraph condenses the entire address in just a few pixels. It’s almost like a mental map of the address. It’s probably what most people would have retained about the address anyways. And then for the topics they care the most they would have a finer granularity view, which is the text itself and the state-by-state analysis of this.
What would you have done differently — conceptually or technically — if you had another go at it?
I would have kept the original visualization, but also would have worked on a real-time visualization that would show the real-time conversational aspect of Twitter.