LoginBayesian Suggestions
Deleted user
Deleted
18 years, 2 months ago at Feb 28 19:07 - If there is one thing to add to listal, it's bayesian suggestions. The group and friend based reccommendations are good, but perfect.
Listal just has to say:
These other listal users have similar tastes to yours.
and
People with similar tastes to you also enjoyed the following things which you do not know about.
This feature is important for the obvious reason of helping people learn about new things they might enjoy. But it has another important use my roommate learned from using this feature on Netflix. It helps data entry immensely.
In the new movie section I want to add every movie I've ever seen. I can't remember that. But I can remember a few. Then the system can recommend some to me, and it is likely that I've seen those also. So now I can rate the ones I've seen and a few more will be reccommended. Eventually after everything I've seen is reccommended to me it will start suggesting things I haven't seen. Much easier than trying to remember every movie I ever saw.
Listal just has to say:
These other listal users have similar tastes to yours.
and
People with similar tastes to you also enjoyed the following things which you do not know about.
This feature is important for the obvious reason of helping people learn about new things they might enjoy. But it has another important use my roommate learned from using this feature on Netflix. It helps data entry immensely.
In the new movie section I want to add every movie I've ever seen. I can't remember that. But I can remember a few. Then the system can recommend some to me, and it is likely that I've seen those also. So now I can rate the ones I've seen and a few more will be reccommended. Eventually after everything I've seen is reccommended to me it will start suggesting things I haven't seen. Much easier than trying to remember every movie I ever saw.
Moderator Admin
Tom
18 years, 2 months ago at Mar 1 16:37 - Yes i would definitely like to add this, I've added it to the list on the main groups page. I'm sure it's not an easy job though I will have to investigate how to program this and possible problems with server load etc..
jotango
17 years, 11 months ago at Jun 14 11:00 - Hi Tom, having programmed something similar for our media trading site ([Link removed - login to see] I can tell you suggestions are really computationally intense. We calculate them in batches on a separate server.
If you generate item-to-item correlations you will have to calculate these recommendations once for each item. If you want to do person-to-person correlations once for each person. But if you want to do personalized recommendations for each user for each item, you need to calculate the product. Our algorithm requires about 0.5 seconds for each intersect, and the product is millions...
Contact me if you'd like more info [Link removed - login to see]
If you generate item-to-item correlations you will have to calculate these recommendations once for each item. If you want to do person-to-person correlations once for each person. But if you want to do personalized recommendations for each user for each item, you need to calculate the product. Our algorithm requires about 0.5 seconds for each intersect, and the product is millions...
Contact me if you'd like more info [Link removed - login to see]