Visualizing Patterns in Receipts

In this example we're examining over 1.8 million receipts taken from a set of hardware stores over a week. For each item sold we have information on its cost, department and class. Each receipt stated what was bought, how many and what was the price. Discount information was also present.

I took this data and constructed a Nicheworks data set by relating items together based on their probability of appearing on the same receipt. I discounted any links with probability factor of under 0.05. The placement algorithm in Nicheworks then grouped the items into component clusters and placed them in a layout that places items close to other items they are often bought with. The results of placing these 40,000+ items are displayed on the next series of pages.

Clicking on images in the main body will give larger versions of the image. Clicking on thumbnails at the end of the page will jump to one of the analysis pages.

gwills@research.bell-labs.com