Confidence in Recommendation

Generating confidence in your recommendations is key to getting people to use them. When using the Music Explaura, you're more likely to trust an artist recommendation that we give if it is an artist you've heard of. Then again, if it is an artist you've heard of, it hasn't actually expanded your musical horizons. The trick is to include enough popular, obvious, recommendations in your results that you can see we're on the right track, then expand into artists you may never have heard of that might match your tastes even better than the ones you have.

I heard an interesting talk during lunch today from the Oracle folks who do retail science. They create forecasts about how products will sell given different factors. To oversimplify, "you'll sell N units of this product if you discount it by 25% starting on this date". One of the non-technical problems they face is building confidence in their forecasts. Sometimes a prediction (i.e. a recommendation) makes sense to the user - have a sale on flowers before Mother's Day and you'll sell more flowers. But sometimes the model predicts something that doesn't have an obvious explanation even through the historical data that is collected indicates that it should be true. The user has to make a judgment call. This, I think, is very similar to the kind of trust a user builds in a recommender system. If the first thing the retail system predicts doesn't make any sense to me (even though it is based on solid facts), my first impression will be that the system isn't going to work. What I really want is for the system to start out by demonstrating that it can predict the "obvious" things (like flowers for Mother's Day, or in our music recommender's case, that I might like Coldplay). Once I'm confident that it can do that, then I'll feel more confident in it when it predicts something less obvious.

On a somewhat unrelated note, one of the speakers observed that a product that has only a few items on the shelf isn't going to sell well. Picture a grocery store where there's only a few boxes of couscous left. Maybe you'll pass on those and get something else. You think "maybe those are the rejects that nobody else wanted". I think it's an interesting look at human behavior that when you turn that into an online store, you can motivate people to buy those boxes by saying "Only 3 boxes left".

Comments:

Hi, trying to contact you about your recommendation service.

I wonder what else I can type just here that will not see this comment wiped out by your anti spiced ham technology?

cheers

Steve

Posted by steve mathers on June 10, 2010 at 02:16 PM EDT #

Got to go visual for an intelligent recommender system.

Posted by bhavesh on April 02, 2011 at 08:45 AM EDT #

Post a Comment:
  • HTML Syntax: NOT allowed
About

Jeff Alexander is a member of the Information Retrieval and Machine Learning group in Oracle Labs.

Search

Categories
Archives
« April 2014
SunMonTueWedThuFriSat
  
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
   
       
Today
Feeds