As far as Mahout is concerned, you just need input of the form "user,item" (no rating necessary) where those are two numerical identifiers. I imagine each logged request contains something like a user ID and other thing you want to recommend -- video ID, item ID, etc. (If it's not numeric, you'd have to hash it and store the mapping, since you do need numeric IDs.)
You would need to use algorithms appropriate for use when there are no ratings, though. Are you thinking of using Hadoop or a non-distributed version? You can do a translation from your logs to a simple CSV format like the above and use that as input. You can also modify the code to read your logs format directly if you like, and avoid the translation step. If you can say more about what you want to do, can probably say more about how to do it. On Sat, May 7, 2011 at 1:40 PM, Shem Cristobal <shem.cristo...@gmail.com> wrote: > Dear All, we are hoping to generate a recommendation from HTTP logs of a > certain web site. Is this even advisable? What sort of recommendations have > you experienced using such HTTP logs? Thanks a lot! > > > > Best regards, > > @shemcristobal >
