You probably want to look at the statistics:
- tagged with X
- check all other Y tags
- count how often Y and X appear at the same time
- divide how often Y appears
- ???
- Profit !!!
As for more information about step 5: this information changes very slowly, so you can cache this material and update it only when you have time.
In the end, what you want is attitude
conditional_probability(X, Y, P)
Which tells you how likely the (P) tag is Y if X. P was calculated in step 4.
Daren thomas
source share