October 18, 2013 at 23:16The mathematical model predicts the outcome of the Kickstarter campaign with a 76% confidence 4 hours after its start
Researchers at the Lausanne Federal Polytechnic School have built a statistical model that predicts the success or failure of a Kickstarter campaign based on funding dynamics and social interactions. Earlier attempts to predict the outcome of co-financing campaigns relied on static indicators, that is, those that were known before the start of the campaign: the size of the amount collected, the availability of video, and topics. Such models achieved an accuracy of 68%.
Swiss scientists collected data on 16,042 campaigns that took place from September 2012 to May 2013. In addition to data on the dynamics of contributions, they monitored Twitter for tweets with the word “kickstarter” and a link to the project page and examined the composition of the project backers - their activity in other projects, typical contributions, etc ... It is the use of a combined model taking into account social interactions that significantly improved the quality of the prediction. 4 hours after the start of the campaign, you can predict its outcome with 76% certainty. After 10% of the time allotted for the campaign, that is, on average after three days, the accuracy increases to 85%.
This methodology has good prospects for a further increase in accuracy - a more thorough analysis of social graphs, the use of other social networks except Twitter. In addition, so far the model gives only the probability of success, but does not name the possible reasons and specific recommendations for improving the campaign.
One of the authors of the methodology, Vincent Eter, launched a site where you can see the result of the model for current projects in real time and download all the source data that the model uses. Scientific work in PDF format with a more detailed description of the model and methodology can be downloaded here .
Swiss scientists collected data on 16,042 campaigns that took place from September 2012 to May 2013. In addition to data on the dynamics of contributions, they monitored Twitter for tweets with the word “kickstarter” and a link to the project page and examined the composition of the project backers - their activity in other projects, typical contributions, etc ... It is the use of a combined model taking into account social interactions that significantly improved the quality of the prediction. 4 hours after the start of the campaign, you can predict its outcome with 76% certainty. After 10% of the time allotted for the campaign, that is, on average after three days, the accuracy increases to 85%.
This methodology has good prospects for a further increase in accuracy - a more thorough analysis of social graphs, the use of other social networks except Twitter. In addition, so far the model gives only the probability of success, but does not name the possible reasons and specific recommendations for improving the campaign.
One of the authors of the methodology, Vincent Eter, launched a site where you can see the result of the model for current projects in real time and download all the source data that the model uses. Scientific work in PDF format with a more detailed description of the model and methodology can be downloaded here .