Increase quality of crowd-sourced info

Question

I'm working on an app that gives traffic alerts in real time and is based on crowd-sourced information. In other words, people use the app and report traffic problems and at the same time they are informed about traffic problems in their area.

A difficult task is how to distinguish real alert reports from fake ones so that the app behaves properly and is useful.

Do you know of any documentation regarding this issue or any programmer stories, insights into this problem? How should this problem be tackled?

What I've come up until now is:

• each person using the app is uniquely identified
• each alert report has a reliability value in an interval 1 .. x
• the reliability of a report is calculated based on the number of users that reported it or confirmed it and the reputation of those people. But how exactly?
• each person has a reputation value which is calculated somehow. But how?

I'm not sure how to handle the reputation/reliability stuff so I'd love some input on this. There must be some documentation on how to create a crowd-sourcing product that works.

Answer 1

Panos Ipeirotis has a fabulous talk on this subject. Yes, you want to incentivize good behavior. If Waze is too complicated this talk will be too, but it will give you a good idea of what is possible if you throw the kitchen sink at it.