GoogleCodeExporter
commented
9 years ago this kind of thing is tricky to get right, so I'm not going to touch it for a
while, but here are some notes.
it's difficult to say "product X is often bought with product Y" because it
depends on how you measure "often".
For example, I'm certain there was a period in the last year where Mentos and
Coke were bought together quite a lot. But let's say that Mentos were not
available to buy anymore - if "often" means "on average, over all time", then
Coke will still be linked to it even years after the link makes no sense.
there needs to be some concrete rules explaining how the relationship should be
decided.
one idea: set up an automatic relationship with a date range (last 6 months).
then for each product, find every paid order that contained it in that period,
and figure the relationships out. Still not easy, though, as you then need to
figure out rules for sales where 2 products of type X were bought and 3 of type
Y - is that counted as 1 relationship?
also, doing this "on-the-fly" would kill the server because of the amount of
computation, so it needs to be done either:
1. partially, each time a sale is completed
2. or in a periodic "cron" event when the server is otherwise not busy
I think #1 is the best option.
you might also want to consider that sometimes marketers would like to "skew"
the results to try create a link between products.Original comment by kae.verens@gmail.com on 8 Nov 2011 at 9:16
Original issue reported on code.google.com by
superstofferon 31 Aug 2011 at 10:38