Re:Georgia Tech Researchers Reveal Phrases that Pay on Kicks...
by minimax on Tuesday January 14, 2014 @02:06PM
(#7058987)
The phrase at the top of the list for phrases signaling that the project won't be funded is "pledged" but the 3rd, 4th and 5th ranked phrases signaling that the project will be funded are "has pledged", "pledged will", and "pledged and". How are you supposed to interpret those seemingly conflicting results?
Another fishy thing is that the top of the list for phrases signaling that the project will be funded is the phrase "project will be". It's fishy because whenever you meet your funding goal, Kickstarter puts a blurb on your page saying "This project will be funded on <deadline>"†. It makes me wonder whether their scraper accidentally picked up some text that was not part of the user created promotional text.
Edit: One more thing. I would be much more convinced of the model accuracy if they had tested it on provably out of sample data. I.e. Take their model as it stands today and use it to generate a prediction for the next 1000 campaigns posted to Kickstarter. Compare the prediction results to the actual outcomes and post the results.
Another fishy thing is that the top of the list for phrases signaling that the project will be funded is the phrase "project will be". It's fishy because whenever you meet your funding goal, Kickstarter puts a blurb on your page saying "This project will be funded on <deadline>"†. It makes me wonder whether their scraper accidentally picked up some text that was not part of the user created promotional text.
† https://www.google.com/#q=%22This+project+will+be+funded%22+...
Edit: One more thing. I would be much more convinced of the model accuracy if they had tested it on provably out of sample data. I.e. Take their model as it stands today and use it to generate a prediction for the next 1000 campaigns posted to Kickstarter. Compare the prediction results to the actual outcomes and post the results.