@ProgramFOX: Are there any guidelines as to when something is a false positive? Everything that is not spam or offensive? If yes, the what about blatantly off-topic stuff, incredebly lazy questions and similar?
@Wrzlprmft Everything that can be added to the Bayesian "good" doctype can be FP'd. If you don't want to hear reports from something anymore, but the post is not worth FP'ing, then execute the "ignore" command. FP = ignore + mark as FP in bayesian db.
You can take the pixels for the current hour and the following hour, and divide by the total. This is sort of assuming that all CMs are equally active in chat over 24 hours, which is probably false.
I think it's simpler and more reliable to look at "talked ... ago".
Ping them at random throughout the day and watch the reaction. :)
I think the two numbers to focus on are the "last talked", which is backward-looking, and the total for this+next hours, which is forward looking.
Since this is just for heuristic purposes, you can take any reasonable combination of the two scores.
Dividing by the total activity in chat seems counterproductive: if someone is generally less active, that makes them a less favorable target for pinging.
Besides, the height of bars is already normalized: the max is 40 pixels.
So, add the bars for the present and next hours: that's a number 0-80.
Then, convert the time since last talked into a similarly-looking number...
E.g., max[0, 80 - (number of minutes since last talked)/3]
This gives 0 for times longer than 80*3 = 240, 4 hours, which probably mean "sleeping".
Add these together, and you have the "availability score", 0-160.
