Automated Game Recaps

[wudu]

I think I might have mentioned this a few years ago here about having automated game recaps put in FOF, whenever the new version comes out. Yahoo Fantasy has implemented a system even more advanced than I expected would come from the technology. And oh man, it's as cool as I thought it would be. Here is an example from my fantasy league:

Fightin' Speed Bumps (2-5, 658.44 points) edged Bulldogs (3-4, 680.36 points) 78.80 - 77.46 in a matchup with several lead changes. They were led by Shonn Greene who had 14.80 points and Heath Miller who had 13.30. Bulldogs was the first to put up points and by Sunday was already out in front 14.10 - 0. They lost that advantage after the early Sunday games and trailed 35.30 - 27.60. After the late Sunday games, the lead changed again, and they went ahead 56.56 - 50.10. However, Fightin' Speed Bumps came out of the Sunday night game ahead by 5.04.

Fightin' Speed Bumps Smooth Moves

Sat TE Jermichael Finley on the bench, where he had the lowest TE score on the team with 3.10 points.
Took the Pittsburgh Steelers Defense out of the starting lineup, which scored 3.00 points on the bench, fewer than the starting DEF.
Shonn Greene scored 14.80 points, which ranked seventh out of all the RBs in the league this week.
Heath Miller had the third-highest TE score in the league this week with 13.30 points.
Fightin' Speed Bumps won despite having only 3 of their 9 starters exceed their projected points.

Bulldogs Regret Tracker

Picked up and started Phil Dawson, who scored a league-worst 11.5% of his projected point total for 1.00 point, the worst of the matchup.
Scott Chandler has now failed to reach his projection for three straight games. This week he scored 1.50 points versus a projected 4.24.
The TE position was a letdown for Bulldogs, where they were outscored 13.30 - 1.50 by Fightin' Speed Bumps.
5 of the 9 starters scored less than their projected points.
Matt Forte fell short of his scoring projection with 10.00 points, just 69.5% of his projected 14.38.

What If

If Fightin' Speed Bumps played Bulldogs every week, they would be 3-4 this season.
Besides Bulldogs, Fightin' Speed Bumps would have defeated four other teams this week.
If they played the same schedule as Bulldogs, Fightin' Speed Bumps would be 2-4.
If Fightin' Speed Bumps had played every team in the league each week, they would be 27-36.
Bulldogs would have lost to six other teams besides Fightin' Speed Bumps this week.
Had they played each other last week, Fightin' Speed Bumps would have lost to Bulldogs 96.20 - 83.06.
Bulldogs would be 3-3 if they played the same schedule as Fightin' Speed Bumps.
Bulldogs would be 27-36 if they played every team every week.

Game Notes

This is the sixth game in a row where Fightin' Speed Bumps has failed to reach their projected points.
With the win, Fightin' Speed Bumps snapped their five-game losing streak.
The margin of victory of 1.34 points was the smallest in the league this week.
Bulldogs got their lowest combined scoring output from the RB position this season with only 21.00 points this week.
Fightin' Speed Bumps improves to 1-1 in close games this season.
The drafted players on Fightin' Speed Bumps put up 75.5% of the team's points this week.
Fightin' Speed Bumps is in the basement in the standings for the fifth week this season.
Bulldogs scored 77.46 points, ranking them eighth in the league in scoring.

Can you imagine that software being applied to FOF box scores? Slap a realistic web interface on your game and suddenly all those complaints about FOF lacking immersion go away.

[Autumn]

There was an article in Wired recently about Narrative Science, a company which produces news articles automatically on things like Little League box scores, or company data reports. Would be perfect for this.

Can an Algorithm Write a Better News Story Than a Human Reporter? | Gadget Lab | Wired.com

[gstelmack]

Ben already generates key headlines for the games on the web.

[aston217]

Yup, it's a longer and more drawn out version of those headlines. The mock draft also has some similar language and is pretty cool for context.

With the right organization it would be easy to develop, I think - just need to decide what game data to look at, and what set of (randomized) messages to generate from it.