Sentiment analysis is hot among financial-market traders, so why not for that other great betting domain, sports?
Serafim Scandalos, an exec at Neurolingo, a Greek natural language processing (NLP) specialist, asked my reactions to Sentibet, currently in beta, which performs text analytics on sports-related tweets. Sentibet looks for Wishes, Feelings, and Predictions in tweets related to specific sports and specific contests/games, producing Sentiment Based Forecast (SBF) for each event.
Serafim gave a lightning talk at an October, 2010 conference I organized. Neurolingo’s Mnemosyne platform seemed promising then, and Sentibet seems on the right track now, worth a look. (I do not have any business relationship with Neuroling although I’m hoping Serafim or a colleague will present Sentibet at my next conference, the May 8 Sentiment Analysis Symposium in New York.)
The Sentibet interface is nice enough although a refocusing on sports contests with available analyses is in order. It’s off-putting to see the “SBF info not available yet!” and “Optional empty text…” messages that now populate many screens. Tell me what you know, not what you don’t.
To get to analyses, try the Most Tweeted and Finished Games areas. (Be patient: Response is far slower than it should be, typically 8 seconds to bring up a page. Something to work on.) Clicking through a particular match…
The Prediction/Feeling/Wish categorization is conceptually quite interesting. It reflects the richness of human emotion, that we may have a variety of different purposes and meanings when we express seemingly similar feelings, attitudes, and opinions. That you hope Manchester United will beat Arsenal doesn’t mean you think a draw won’t be the outcome.
Similarly, I really like the Team A (Home)/Team B (Away)/Draw categorization. It’s goal-aligned and therefore far more useful than the positive/negative/neutral sentiment classification that’s typical of the Twitter-sentiment toys that are far too common.
The best part is that you can view the tweets behind the dual classifications, and you can apply filters to select only tweets in certain categories. I was skeptical of the analysis displayed: 2% of tweets on an Arsenal-Manchester United match Wished for a draw?! (Who hopes for a draw?) Well, SentiBet got it right. Witness:
@darrenpage1983 revenge weekend spurs bt man city or atleast don’t get stuffed 5-1 arsenal man u would like to c a draw there as never want the scum to win
@Slickon_CFC Good Morning guys. Its super sunday today and i’m hoping for a City win and a draw between United and Arsenal. #Top3
@annemacey02 @Betherz_BCFC I want my spurs to thrash manc and a manu arsenal draw :p
Note a few points:
- Accurate co-reference that correctly sees “Manchester United” – “man u” – “manu” as one and the same.
- Ability to distinguish two different matches in a single tweet.
This is strong text analytics.
It’s a good sign when an app leaves you hungry for more, for more data, for additional analysis. Sentibet does. Within the Finished Games section, it would be interesting to be able to correlate scores and forecasts, to understand the correlation between tweet aggregates and match outcomes. How well do Sentibet forecasts predict final scores? It’s a Neurolingo work-in-progress as explained on the Sentibet Post Game Analysis page.
It would be even cooler to connect forecasts and betting lines, that is, bookmakers’ odds. As the expression goes, “The proof of the pudding is in the eating.” If you consider sports betting as a business, they payoff isn’t in scores predicted, it’s in your gambling winnings. Financial-market traders exploit the price-expectation gaps, and so do winning gamblers. Maybe another to-do for Neurolingo?
Serafim Scandalos tells me Neurolingo is seeking financing to accelerate service development, also that Sentibet is simply one application of the underlying Mnemosyne platform, meant to demonstrate possibilities to potential customers and investors. Judging from Sentibet, Neurolingo seems like a good bet.