September 30, 2006 9:25 pm by Markus.

I just returned from the ECML Data Mining Workshop and one talk I found particularly interesting. In the talk Network-based marketing: Identifying likely adopters via consumer networks (S. Hill, F. Provost and C. Volinsky) presenter reported on a successfull marketing campaign. Rough summary from the talk: A phone company was introducing a new service and from past experience they had twenty-something marketing segments for people that were likely to buy that they would write to, call or otherwise inform of the new service. Since the phone company has access to the call records they extracted a list of friends these likely buyers frequently call and started marketing to these people as well. The cool part is that the response rate from the friends was about 3 times higher than the likely-buyers response rate (or was it even the buy-rate). That said, so many companies now started to collect (or have available to them) social networks data, e.g. Skype (now EBay), Google (GMail invites), MySpace, Facebook etc. Most likely this will change the ways of advertising quite a bit. Sidenote: the companys lawyers felt this is legal, because the company owns that call data. Interesting how this is legaly different from the NSA-survailance-program the US government has been doing.

Posted in Data Mining | Print | No Comments »

September 17, 2006 6:23 pm by Markus.

I just got back from a brief stay in Las Vegas. I didn’t gamble, but got thinking about the game of Roulette. Obviously all the games that are played rely on unpredictability in one way or another and require real randomness, just like many cryptography applications such as SSL. If the state of, say the deck of cards in a black jack game would be known, the outcome becomes predictable (see also: cracking the Netscape SSL random number generator). The black jack card counting thing is getting old (and I noticed that they cut the deck of cards and throw away a number of cards - which leaves the gambles with no information what cards are now in the deck), but recently I read about people using a laser scanner to predict the roulette wheel outcome. An interesting bit I noticed was how Roulette is played differently here in the US. Besides that there are two zeros on the table (giving the bank an even greater chance to win) the croupier changes the spin of the wheel (he/she stops the wheel and starts spinning it again) and puts the little ball in after all the bets have been made. I recall that in Europe they spin the wheel and put the ball in and then allow for bets being made for a certain amount of time. That certainly makes it impossible to predict anything at all at the time of betting (unless the wheel would be biased somehow, say by a slight tilt). Also the displays that show the last couple of numbers seem to malfunction occasionally and display the wrong numbers. I guess the only way to win in a casino would be if any of their “random number generators” were unbeknownst to them were not truly random - but then again they won’t like it if people win.

Funny sidenote… you can find amazing, “infallable” Roulette Systems on the web for only $27 Then there others that are available for free. For example, the pivot system. The inventor claims that :”It is a fact that numbers on a roulette wheel tend to repeat often.” Some of them might even work (play simple chances, double your bet everytime you loose) assuming that there would be no limit at the table (and ignoring that the house still has an edge).

Posted in Statistics, Ramblings, Random | Print | No Comments »