Kickstarter Data Project

Transcript

1. KICKSTArTER SUCCEss
2. Our Data. We Had 87000 projects. 300,000 users. We Used 15000 projects 400,000 users (with repetitions). We Collected Twitter + Facebook Shares. Project Creator’s Facebook Friend Count.
3. Cleansing. We Were Left With 12000 projects. We Chose Projects with >1 backers. Log scale. We Discarded Facebook data (extremely inconsistent)
4. Hypothesis. Global reach affects the success of projects.
5. Backers. Hexbin Density Plot
6. Projects. Hexbin Density Plot
7. Features. Backer Location Variance Presence of Video Backers Comments Images Pledges Minimum Pledge Maximum Pledge Tweets Facebook Friends Art Dance Design Fashion Film/Video Food Games Music Publishing Technology
8. Backers Comments Pledges Updates
10. Significant Features. Backer Location Variance 2e-16 Has video 0.000455 Backers 2e-16 Comments 2e-16 Image Count 0.000290 Maximum Pledge 2e-16 Tweet Count 3.75e-05 Design 3.60e-05 Games 3.25e-14 Technology 0.001938
11. (without Backers) Comics 0.049473 Design 3.60e-05 Food 0.013668 Games 3.25e-14 Technology 0.001938 Backer Location Variance 0.001512 Comments 2e-16 Image Count 0.000524 Number of Pledges 0.017430 Maximum Pledge 2e-16 Tweet Count 9.14e-06 Significant Features.
12. 1504 130 78 3033 Classifiers. Random Forest 94% Accuracy
13. Classifiers. 0 0 1634 3111 3052 1516 59 118 GaussianRBF 65% Accuracy Linear 96.3% Accuracy SVM
14. Density plot: Hidden Radio and Bluetooth Speaker