Word based predictive model question/theory

[smartbets]

Has anyone ever attempted to build, heard of, and or used some sort of data model that analyzes the spoken and or written words of a particular sports organization in the days leading up to a game to try and get a probability percentage on their expected level of energy/effort/motivation for that particular game?

What I was visualizing/thinking about was some type of database/model that stores all of the written and audio interviews/statements from the coaches and players from a sports team after a game has been completed and leading up to the start of the next game.

One would then analyze the words used, and if possible, the context of the words and sentences to look for any patterns that might infer an increased or decreased level of play from that team for that game.

I am referring to the words that have been identified as "power" words and or "limiting" words by studies done and by the various fields of social science that try to quantify words and how they affect our behavior(like NLP). I have seen many lists of words that suggest forward positive momentum generating behavior, and words that infer declining or negative behavior related to the specific activity. I am wondering if there is a way to detect a higher level or lower level of either of these types of words, and if there is some statistical significance to their usage compared to previous weeks, if that would have any effect on the level of performance for the upcoming game.

For instance, saying "we have to hang onto the football" is different from "we can't keep turning the ball over". Or "we will bounce back from this defeat" is different than "we learned from this game and are ready for the next game". "We didn't get enough pressure on the QB" versus "they did a good job of picking up our blitzes". "We can't get away from the running game" is not the same as "we need to run the ball more", etc.

Obviously, like many research projects, this could be completely worthless information or may lead to insight into assigning probabilities to the expected level of play from one or both teams competing against each other.

I realize that this concept is very outside the box, so I would hope feedback I get on this would be constructive.

Thanks

[runnershane14]

excellent idea in my opinion..but i find it hard for one individual to take on such a monster of a project..good luck to you

[samserif]

It's an interesting idea. I'm guessing it was inspired by this article on using similar methods to predict how college QBs will perform in the NFL, right?

The biggest problem I see is the scarcity of data. There's more data at the pro ranks than at the college ranks but it's still pretty sparse. Then you've got to sort through a variety of individuals to find the ones who are most predictive of team performance on a week by week basis. Skip the coaches who are good at (polished) public speeches, the prima donna players who don't reflect the team as a whole, the leaders who always appear upbeat, etc.

Like I said, it's an interesting idea, but I'm skeptical of the value given the effort.

[princecharles]

Someone like you is where the next level of capping will come from.
I applaud your thinking.
With this particular idea I feel too many distractive and poorly refracted pieces of communication abound.
Also I would need the private communications, (lockerrooms, practices, one on one coach to player) to feel like I was procuring the purer/cleaner/unfiltered communication.
Nice job though, right in my wheelhouse of interest.

[YouMama]

ya, awesome Idea

[peacebyinches]

I too really like this idea, I would be willing to help and brainstorm if you want. Just message me.

[smartbets]

thanks guys - like a few of you said, def a big project

sam - great article! i never read that article, but thanks for sharing - looks like there are some sources cited in that article, and companies are already doing this - very interesting stuff

[byronbb]

They do this for financial markets. I thought Bloomberg Sports was basically this idea but it seems only tailored to MLB atm.

[evo34]

http://www.sntmnt.com/ is an example in finance. Personally, I think it's a poor use of resources to try to develop such a strategy. EVen if you created something that appeared to be predictive, social media and sports reporting content styles are changing all the time, so you'd be backfitting to a moving target -- which to me is a recipe for disaster.

[Wrecktangle]

Maybe fun to think of, but crowd-sourcing is not worth the effort in the NFL, at least.

[samserif]

http://www.sntmnt.com/ is an example in finance. Personally, I think it's a poor use of resources to try to develop such a strategy. EVen if you created something that appeared to be predictive, social media and sports reporting content styles are changing all the time, so you'd be backfitting to a moving target -- which to me is a recipe for disaster.

This company relies on measuring investor sentiment, which may be a good predictor of stock prices but isn't a good predictor of scores. What you need is a way to gauge team sentiment, which is why interviews are plausible (though I don't think it'll work well for reasons I already explained). On the other hand, this "team sentiment" approach wouldn't work in finance because verbal and written statements coming from public companies are highly crafted and reveal nothing that hasn't already made it into the news.

[wrongturn]

Just read an article that Alabama head coach was unusually relax and confident during interviews before the rematch against LSU in title game. Wish I had noticed earlier.

[evo34]

This company relies on measuring investor sentiment, which may be a good predictor of stock prices but isn't a good predictor of scores. What you need is a way to gauge team sentiment, which is why interviews are plausible (though I don't think it'll work well for reasons I already explained). On the other hand, this "team sentiment" approach wouldn't work in finance because verbal and written statements coming from public companies are highly crafted and reveal nothing that hasn't already made it into the news.

My point is that parsing published information in an effort to find hidden edges in predicting future behavior is not a good use of talent or time. Stocks, sports or anything else.

[samserif]

My point is that parsing published information in an effort to find hidden edges in predicting future behavior is not a good use of talent or time. Stocks, sports or anything else.

Yep, got it. And to your point that it's a moving target, I can easily see a team banning unsupervised interviews or forcing coaches and players to attend PR classes after the first time someone actually has some sort of success with a strategy like this. As soon as it shows value, it goes away.

[GreyMatterStats]

It would take a monumental effort, but the concept if fascinating.

I love reading and thinking about ideas such as this.

Thanks for sharing your thoughts.