Sunday, December 21, 2008

Announcing: Bubble Analytics

This is a double-release along with the bubble time-machine which has been added to the bubble search functionality. This entry focuses on the "Analytics" link on certain bubbles.

Analytics are graphs of certain bubbles that I have rendered out... for instance... "president", or "crisis", or "new orleans". The graphing system counts up the 30 most related bubbles for each day back in time from today until the day that you specify as the beginning. Then the system makes a graph of the context of that bubble. You can then choose exactly what you want to be graphed out in the context from your graph options on the right.



An example graph: The President Context Graph - click the link or the graph to actually play with the graph, and all the different things that cross paths with the word "president".

Now, looking at this graph, you can see "Mccain" in red, "Obama" in blue, and "Bush" in orange... the other two plots are of "Debate" in green, and "Election" in purple.

Here are some things you can notice when you examine the graph further. Keep in mind that this is solely graphing the list of most related bubbles over time, meaning that the higher up on the graph, the more related... or the more times it's mentioned along with the context bubble, (in this case "president"). What you'll find in the president context graph:
  • In the next 30 days following october 27, the time of the election (marked on the graph in pruple), "obama" overtakes "bush" in relativity to the word "president"
  • Looking at the whole year, the word "mccain" only is slightly more related once, in the same 30 day period as the debate. The rest of the year, obama dominates the relation to the word "president" in news media.
  • The election got a good amount of context relatedness to the word "president" throughout the year, and had a peak in november, while the word "debate" had it's peak contextual closeness in october. Thes are when these two events actually happened.
The system is still being worked on, and improvements made to the interface on an ongoing basis to make things easier to control. If you want to have a new context brought forth in the graphing system, you should get a newsbubbles account, and then request one from me by e-mail. For now, you can choose from the contexts that I've already pre-rendered and you can use the system to graph any of the bubbles that show up in the context over time. There are literally millions of plots you can make with the contexts I've provided you.

In the future I will be offering analytical services to companies wishing to compare these graphs to market information.

More later...