WinterMilestone 3 MarkWhiting ReputationIdea:Defrosting Cold Start
The cold start problem, giving good tasks to new users when we don't have data on them, might be solved in a way reminiscent to how we solve it in cars; by running them for a few minutes before driving when its very cold out.
In this domain, I think this would involve giving a new user a few tasks to work on to get used to how the system works. At the same time, we might start to use their responses to better match them with good tasks, give them instructions on how the platform works and how some users choose to optimize their experience, and most critically, start evaluating key signals for a reputation engine. This warming period might involve doing a few different types of tasks that we have specifically designed to help us work out what kind of workers or requestors they might be.
- Help new users understand the system
- Give new users a good model of the incentives and performance factors of the system through experiential learning.
- Help the system get a sense of new users abilities and relevant behaviors
- Use carefully designed sample tasks and other activities to understand what a user might be like. As the system grows this data could become very precise.