Every couple months on A&E or Discovery they have a show on ‘Inside the Casino’ where they showcase the technology at use inside the Las Vegas casinos. Jeff Jonas is someone who helped build that some of the tech. Specifically Non Obvious Relationship Analysis (NORA) which seems to be able to be simplified to ‘really cool data mining’.

Take for instance the casino problem: 2000 cameras on the floor, 50 monitors in the security room, and 3 people watching. How do you know which monitor to watch? And more daunting is which camera to show on the monitor in the first place?

This data overload contributes to what he calls ‘enterprise amnesia’ – data a is in one part of the organization and data b is in a different part, but they are not linked in useful ways preventing the interesting questions from being answered? Take for instance the addresses of your dealers and those of banned players. The casinos have that information already, but not necessarily correlating it. He also cites a (admittedly small) percentage of employees in retail have been previously been charged with theft from their now employer. The information is there, but not in a useful way.

Both those examples could be done through traditional data mining techniques. Where his stuff goes is one (or two) steps further. In the systems he works with the data is only part of what is stored in the database. The queries are too. This means

  • The data finds the data
  • The relevance finds the user
  • Queries find other queries (which leads to collaboration between the people who are interested in the same thing as they are searching for it in the data set)

These techniques seem to me that they would be well employed in some of the central QA systems, especially bug trackers.

Most systems these days lets you save queries / filters so you can get information you care about, but that is only the first part of the solution. The next part is to have those queries run all the time and have the data reach the user. RSS is ideal for this and products like FogBugz incorporate this pushing of result sets to users. The final part of this is to have the system let you know who else was interested in this new bug as a result of their filters.

This might seem overkill for small, collocated companies where everyone generally knows what the others care about, but I could see this being massively useful in the HPs or Motorolas or Microsofts of the world.

You can listen to the full podcast (only about 20 minutes) at it’s IT Conversations page.