Did This Get Done?

Predicting user behavior on task management software

Transcript

1. Did This Get Done? Predicting user behavior on task management software Eric Van Oeveren
2. “Transforming Teamwork”
3. “Transforming Teamwork” Plans
4. “Transforming Teamwork” Tasks
5. Tasks: marked as complete “Transforming Teamwork” Plans: updated with percentage
6. Some things don’t get done. Can they be predicted? “Transforming Teamwork”
7. Features: - Task/Plan - User events/day - User survey scores - Group survey scores Random Forest
8. Random Forest Results - Accuracy: 83% - Precision: .78 - Recall: .67 Most Important feature: task/plan classification
9. Tasks are reliably updated, but Plans are not
10. Tasks are reliably updated, but Plans are not
11. The model does predict which items users mark as finished Tasks are reliably updated, but Plans are not The model does not predict which items users finish.
12. Features: - Task/Plan - User events/day - User survey scores - Group survey scores Random Forest Plus Measures of User Engagement: - Updates in 30 days - Description Length
13. Results Accuracy: 83% → 86% Precision: .78 → .79 Recall: .67 → .79
14. Deliverable: Model that flags items unlikely to be updated CrossLead intervenes (e.g. send reminder email) Items get updated Client stays organized, CrossLead gets better data
15. Eric Van Oeveren PhD Candidate - Physics