Friday, January 22, 2010

2010-01-22: Conference Championships

It is time for the Conference Championship games. Both sets of teams have worked hard to make it this far but only two can go to the Superbowl. This week all of our Data Mining algorithms are in agreement as to who the winners should be.





For the AFC Championship, Indianapolis hosts the Jets at 3:00PM EST. The Colts with Peyton Manning have arguably one of the best pass offenses in the League however their rushing game leaves a little to be desired. The Jets have one of the best defenses and their pass defense has given up less yards than just about every other team over they year. The algorithms call for the Colts to win but if I was a betting man I would wager that even with a loss that the Jets would cover the spread.




For the NFC Championship, the New Orleans Saints host the Vikings at 6:40 PM EST. The Saints have had a phenomenal offense this year with some explosive plays. Their defense however has been lackluster. The defense seems to be better at forcing turnovers that actually defending. Although the metrics indicate that turnovers are very important in winning a game. I just don't see the Vikings pulling this off. Vegas has the Saints - Vikings much closer than Colts - Jets, but our algorithms have the Saints beating the Vikings by a higher percentage than the Colts beating Jets.

The Superbowl represents the culmination of a life-long dream for many of the participants and regardless of the outcome this should be an exciting weekend for all football fans.

-- Greg Szalkowski

Friday, January 15, 2010

2010-01-15: Divisional Playoffs

Dallas has really taken off and seems to be peaking at just the right time. The Cowboys decimated the Eagles for a second time and they almost made it look easy. While our research seems to indicate that teams with good offensive passing generally do better, the recent Dallas surge lends credence to the old saw about the run game. According to our numbers Dallas has the best run offense this year and close to the best run defense. The Dallas pass statistics are not that bad but they are overshadowed by the likes of New Orleans and the Colts.

The Colts bring up an issue that I alluded to last week that I had to deal with before running this weeks algorithms. Because Indianapolis sat their starters for the past few weeks I need to decide whether or not to exclude those weeks from the calculations. If I include weeks 16 and 17 for Indy then Baltimore comes out as the predicted winner. However if I ignore those weeks then the Colts are the pick.

We felt that that those two weeks are not indicative of the true performance of the Colts during the playoffs so we dropped Indianapolis's performance numbers for those weeks and ran the algorithms and this is what we came up with.





HomeVisitorSVMPagerankVote
NOARINONONO
INDBALINDINDIND
MINDALDALDALMIN
NYJSDSDSDSD


--Greg Szalkowski

Wednesday, January 13, 2010

2010-01-13: JCDL 2009 Doctoral Consortium Abstracts Published

The Winter 2009 Bulletin of the IEEE Technical Committee on Digital Libraries (TCDL) has the extended abstracts from the Doctoral Consortium of the 2009 ACM/IEEE Joint Conference on Digital Libraries (JCDL 2009). Thanks to Lisa Spiro for such a great job putting together the latest IEEE TCDL Bulletin.

Megan Winget (UT Austin) and I organized the doctoral consortium this year. We were fortunate enough to have 24 submissions this year -- a record number -- of which we selected 14 for presentation. Those 14 students had their extended abstracts reviewed by a 10 person committee prior to the consortium and then they presented their research to the committee. More information about the participants and processes is available in our opening editorial and the JCDL 2009 Doctoral Consortium page.

Neither Megan nor I will be involved in the 2010 JCDL Doctoral Consortium, to be held in Gold Coast, Australia. Those wishing to participate should check the JCDL 2010 web site for details.

-- Michael

Tuesday, January 12, 2010

2010-01-12: Classes for Spring 2010

The Web Science and Digital Libraries Research Group is offering two classes this spring:

Please note that there has been a change in instructor for 495/595: Martin Klein is now the instructor for Spring 2010. Although the instructor has changed, the class will be similar to the previous offerings of the course.

Michael Nelson is still the instructor for 751/851.

--Michael

Friday, January 8, 2010

2010-01-08: NFL Playoffs Wildcard Weekend


The NFL playoffs have arrived so grab some snacks and find a good spot to watch the games. We are going to have our own playoffs here by pitting the best performing algorithms from the regular season against each other to see which one does the best job predicting the playoffs and superbowl.

While I was running the algorithms I contemplated how New England was going to perform without Welker and especially after reading that Brady has been playing with broken fingers on his throwing hand. How can my algorithms take things like this into account to improve the accuracy.

Having the Colts sandbag the last few games of the regular season by benching their starters already throws a monkey wrench into the works. Then there was the Dallas-Philly game, Philadephia had the playoff spot already so did they really take it easy and let Dallas stomp all over them like that last week or are they in for another beating? I am sure Frank would say that Dallas really is that good. If their defense can keep it together they might be able to pull it off.

For our playoff predictions we used three algorithms. All three algorithms use only box score data for training. They are a Support Vector Model, Google's PageRank, and a third model that bagged all of the predictors by giving each predictor a vote for the winner and the teams with the most votes is the predicted winner.

Enough talk the algorithms have finished running. Without further ado here are our predictions for the wildcard games.





HomeVisitorSVMPagerankVote
CINNYJCINNYJCIN
NEBALNENENE
ARIGBGBGBGB
DALPHIDALDALDAL


--Greg Szalkowski