As you probably noticed, we recently
added insights to all of our questions at Predictify: simple text phrases which
describe an interesting piece of information about which predictors are
predicting what. We have developed a data synthesis system which computes simple,
easy to understand insights into user predictions.
Grinding through information about
predictors and their predictions, a specially designed data synthesis system finds
interesting facts about the data and generates a text phrase, and a graph
corresponding to the phrase. For example, on the question
"Will there be a terrorist attack on U.S. soil in 2009?" an interesting insight generated was “20% less beginners predicted No than
scholars” which suggests that more accurate predictors are also more cool-headed
about terrorist attacks in the U.S.
An
important piece of this system is user feedback. As insights are clicked on,
read, and quoted, their quality rating will increase, and components from the
highest quality insights will be combined to create more, increasingly meaningful,
intriguing, and perhaps entertaining, insights about predictor’s opinions. This
means that each time you check back on the insights page for a question, you
will probably see new, more interesting insights than before!
For
example, the terrorist attack question might also h
ave the insight “10% more men
predicted Yes than women”, and this insight and “20% less beginners predicted No
than scholars” might be combined to produce “10% more women predicted No than men”
by taking the demographic, gender, from the first insight, and the prediction, “No”,
from the second insight.
After
repeated recombination of insights (with some random noise thrown in there as
well), you may discover some very intriguing things about different group’s
predictions. So, enjoy digging deep into the insights, and feel free to send us
feedback, comments, and suggestions about Predictify Insights!
Charles