It’s finally here. Super Bowl XLVI. In just a few hours from now, The New York Giants will face off the New England Patriots in the culmination of the 2011-12 season and playoffs. This is television’s biggest event and the sports pundits as well as arm chair quarterbacks are already off making predictions on who will win the Super Bowl this year.
oObly, however, turned to sentiment analysis scientists and asked them what they can come up with. Saygent, a voice-sentiment analysis startup in Silicon Valley, California, is predicting that the New England Patriot, led by quarter back Tom Brady, will win the Super Bowl defeating the NY Giants with a final score of 26 – 23.
How Did Saygent Arrive At This Winner Prediction?
In a traditional poll, a bunch of people are asked to predict the score, and then the results are averaged out for an overall ‘crowd’ prediction. This approach suffers from 2 major flaws.
- Every prediction is treated equally. So a football experts prediction, an arm chair Quater Back’s prediction and a soccer mom’s prediction all get mixed into one giant pool of responses with equal weightage.
- Biases skew the results, because most people rarely vote against their home team. For instance, I always predict a New England Patriots win regardless of how badly they are doing in a given season.
To avoid these flaws, Saygent applies science and algorithms to arrive at its predictions. In stead of a simple online poll, Saygent chose to conduct voice surveys to get crowd predictions. While this is far more challenging than conducting a simple text poll, the effort is well worth it because a voice-sentiment analysis engine can get so much more data than a simple score.
For instance, the Saygent analysis engine was able to determine implicit and explicit biases based on how people talk about each team. Someone could declare a team as their home team or even they use strong words to describe how they feel about a rival team.
By analyzing the way people talk about the game the analysis engine is able to infer who is actually knowledgeable about the game and the teams and who is blindly taking a stab in the dark. This in turn, ensures a more accurate data sampling.
For this Super Bowl prediction, Saygent had people call in to their voice surveys, and then ran their voice-sentiment analysis engine against the entire set of data. All told, 205 people were voice-surveyed. The saygent algorithm then eliminated nearly half of the responses (105 to be precise) due to what they call their ‘Gold Standard’ and included the other half (100 surveys) into their analysis. Surprisingly, nearly half of the people surveyed were women.
What is Sentiment Analysis?
Sentiment analysis is not new. It has been applied to tweets in the past, to understand emotions and how people feel about a certain brand or trending topic. A group of students from Stanford University previously applied NLP (Natural Language Processing) and came up with a twitter sentiment analysis site. They even used it to successfully predict results of local election in India.
Saygent, however is the first company that I know of, that is applying sentiment analysis to voice / speech.