Dynamic Visualization of Conference tweets
I have been analyzing/gathering tweets from conferences. Sometime back I saw tweets from a conference which had a lot of healthy interaction. In this post I attempt to visualize the interaction among participants of a conference based on their tweets.
Note that the focus in this post is more on the visualization technique. It is not on reaching some definite conclusions about social networks – there is a lot of data to analyze and the data needs cleaning up before you can make definite inferences.
Some of these graphics are complex and pretty. It is important not to let that influence your opinion of the value they convey.
I used d3js for visualization. I have substituted the original names with fictitious names in the visualization. Remember to click on the images to view the actual visualization.
Visualizing all tweets
I started by visualizing all the interactions in a conference. This produced the hairball shown below. You can move your mouse over the chords to see the influencers and to see who they interact with. I then created a visualization of another conference to compare. The first conference had roughly double the tweets and it may not be a fair comparison. I am not sure if this is good enough to conclude that the first conference had more or better interaction.
I then used another technique, a matrix diagram (also called adjacency matrix), to visualize tweets (from Mike Bostock’s post). When you view the conferences in the matrix diagram the differences are much more visible (you may have to zoom in using the browser). The first conference has more tweets. There are also more tweets of a darker colour – darkness indicates a larger number of tweets. (Note that this could be a symptom of the size of the conferences.)
You can extend the visualization to other subsets of the data. I created some for tweets which were replies. This is equivalent of a smaller subset of the original data. As with the entire dataset, the matrix diagram shows a clearer view of the interaction. The chord diagram is prettier, but it’s difficult to characterize the entire set of interactions by looking at the diagram.
You could extend the analysis to a subset of participants, e.g., based on age/experience.
Notes about the visualizations
In both diagrams tooltips show the number of tweets. You should be able to figure out the tweets to and from a pair of participants. The chord diagram is better to see all the activities of a single participant. The matrix diagram is better for a view of the entire set of communications. The matrix is also good if you want to see the interactions between a specific pair of participants.
If you are interested in performing similar analysis for a social network and need some help feel free to let me know.