On 08/06/14 21:28, Rob Nickerson wrote:
Measuring the success of a bot can be done in the same way that the
success is measured in a statistical model; Type I and Type II errors.
So for example, if the test is to attach wikidata tags to churches then:
* A type I error would be failing to add the wikidata to the associated
church (we can accept large errors in this case as it represents the
current no-bot situation)
* A type II error would be adding the wrong wikidata to a church. We
need to minimise these. I'll suggest a 1-2% error as this matches the
general quality of public data.
I was taking the question of accuracy as a given.
My point was not whether we can accurately add the right tags, but
rather whether doing so is actually useful.
There's an almost infinite number of things we could add, but if nobody
ever uses them, is the time take do so wasted?
Tom
--
Tom Hughes ([email protected])
http://compton.nu/
_______________________________________________
Talk-GB mailing list
[email protected]
https://lists.openstreetmap.org/listinfo/talk-gb
