Login
Computer Science @ Rochester
Friday, September 23, 2011
12:45 PM
CSB 703
Naushad UzZaman
University of Rochester
TwitterPaul: Retrieving, Ranking and Aggregating Twitter Predictions
In this work we explore the use of social media for game prediction. We prototype our system on World Cup Soccer 2010 tournament with Twitter data. We approach the problem in two different ways. In the first part, we focus on prediction retrieval where we retrieve the predictions for a game and rank the predictions in a meaningful way. We approach this problem as an exploratory search, where the user can explore prediction tweets, compare the expert opinion (users with higher previous success) against all predictions (crowd voice) and the prediction tool outcome, and find successful predictors. We conclude this part with suggestions for better prediction ranking for the prediction retrieval problem.

In the next part, we use the extracted predictions to predict the outcome of the game. We compare our results to strong baselines and against the betting line (prediction market). We found that quality extraction in quantity matters more than just the very large quantity and our performance is close to the betting line. However, when we experiment considering the previous history of the predictors, we do not find any significant difference. The history of predictors doesn’t hurt the performance though, so we could consider it to discard the spammers. We believe our domain independent framework can be used to predict other sports, elections, product release date and other future events that people talk about in the social media and has an advantage over other methods in that we can discard spammers by checking previous history.