Yeah it needs a few things done before it can become usable. Get rid of the assumption knowledge = frequency of keyword. Normalize the data. Account for population differences.
Would also be nice if you could pan around but it looks like the code relies on a static image.