The Forces of Twitter

Interactive Data Visualization

Staticians have found that Fox News, Krassenstein, CNN, and theHill are the top 4 most "influential" media outlets on twitter. What does that really mean, though? I believe it means that they are influential in swaying public opinion and informing people on the news. However, the means through which this is done is frequently through anger and scare tactics. It isn’t so much about reporting the news as it is telling you how to feel about it. This causes the replies to the twitter feeds of these media outlets, as well as any other political figure on twitter, to be fire of online debate and unhappiness. A political figure will tweet something, and below the left and the right will fight the culture war. Rarely does the original tweeter get involved, but when they do it gets ugly.

Inspired by this hatred, I gathered online data manually from Twitter, and organized it into two variables. One variable is a calculation of replies/likes to threads replying to a tweet. In other words, the ratio of people talking not to the media sources I gathered data from, but rather arguing with each other in the replies of those media sources. This is done because online arguments do not occur so much with the person who posts something, but rather with other people who are replying to the same thing. There is a phenomena known as getting “ratio’d” which is when there are far more replies to a tweet than likes, meaning that people do not like what you have to say at all.

The other variable I collected is a word count. In general, longer tweets are more argumentative or aggressive, but this is not always true. Sometimes an image with no words can set people off. Other times, a long tweet may be really positive and non-controversial. But, as a general metric for the kind of tweet being tweeted, the word count is useful.

After I collected this data, I created a program in Processing (seen in the video above) to visualize it. First, I created a “menu” with four particle systems that would represent the four most “influential” media outlets. Then, I applied movement and saturation variables to those particles to have them represent individual tweets.

The ratio data is applied to movement in a display state. The goal here is to provide a way of showing the aggression of twitter by having objects overshoot movements based on those ratio vectors applied to them. What does that mean? Well when the particles are moving up and down, there is a line in the code that tells them to switch the direction of the force being applied to them, so that they go down. But, if a particle has a lot of inertia, that reverse is going to take a while, and it will overshoot that line. Particles that are moving off screen and fast are really stirring the pot, getting a lot of replies, and fewer likes.

As for the word count, that was applied to saturation. It can be seen based on color what kinds of tweets stir the pot more. Sometimes a low saturated (fewer words) tweet sets people off, while in other cases a long argument does.

What begins to happen as you watch the data visualization is that you begin to see the communities around each of these media outlets. Fox News, for example has a community that has no in between. Tweets either set people off, or make no impact at all. Without seeing the data move in this way, it would be much more difficult to see that conclusion.

Currently, I am working on adapting the code to Openprocessing.org so that it can run in your browser. I am also working on a live version that can make the calculations and get the word counts live from any twitter account using the API. When either of those projects finish, this page will be updated.

If you would like a copy of the code for your own adaptation or use, feel free to email me.