Automatic wrong site suggestion when posting?

Question

What do people think of the idea that as soon as a question or question title contains certain key words (perhaps a volume of key words), in much the same way it suggests that your post might be subjective, it suggests that perhaps you're on the wrong site?

E.g. on ServerFault, every time someone uses the words "Home Network" or something similar, it popups up a little box "If your question is relating to a home PC or a home network, it belongs over at Super User." (obviously the phrasing needs to be worked on).

Or on any of the sites, if the question contains the name of the site (e.g. Super User Colour Scheme hurts my eyes), suggest the Meta site.

This would obviously catch the question before it gets voted to close 5 times (which on SF can take quite a while, as there's only about 35 3k users, and obviously even fewer on SU).

Thoughts? Ideas?

_{Prepares for a flurry of downvotes and "Exact Dupe" even though I can't find the dupe}

Answer 1

I think sometimes as programmers we get obsessed with programmatic solutions to the detriment of practical ones.

Tiny fractional slivers of human effort are pretty darn effective at this already.

It would take a substantial engineering effort to even attempt this, and it would likely cause more problems in false positives and bugs than it solves.

Answer 2

Perhaps the 'suggest' feature could search all SOFU sites and if the results come disproportionately from one (80%?) it could be suggested to go there instead.

Answer 3

It's a great idea to suggest to users that they may be posting into the wrong category!

But rather than doing that with a text matching algorithm, it sounds like this is a classic application for a Bayesian classifying algorithm:

• http://en.wikipedia.org/wiki/Bayesian_spam_filtering

Basically we have four datasets: the existing databases of posts for SO/SU/SF and Meta. Every post has already been "classified" - we can be reasonably sure that the classification for each post is correct if it has a certain number of views and/or upvotes.

So using that dataset of classified posts, we should be able to use classic Bayesian filtering to determine which bucket/category each new post should be posted into. This is how Bayesian algorithms filter emails... using Bayesian algorithms run on existing datasets, they determine if each new post should be classified as Ham or Spam. In this case, our Bayesian filtering algorithm would determine if each post should be in SO, SU, SF or Meta.

Something like this would be adaptive - as more sites are added, the Bayesian algorithm would learn which site to sort each post into. This would scale nicely - rather than having to brainstorm new text-matching patters for each new site, the Bayesian algorithms would find out what words tend to appear in posts on that site and pattern match accordingly.

The StackExchange engine could use it as well, to suggest an appropriate site in the event that a user posts an offtopic post into one of the SE-powered sites.