Recommending already consumed items

[Mario Levitin <ma...@gmail.com>]

Hi all,
In some recommender applications the system might recommend already
consumed items. For example, a hotel recommendation site might recommend
 hotel A to a user who already stayed at hotel A before.

In order to recommend already consumed items we have to rank all of the
items (consumed and unconsumed ones)

My problem is how can we rank all the items? Let me explain.

If ratings are available I think the problem is easily solved. We can rank
all the items because we have both the ratings given by the user and the
predicted ratings.

However, what should we do if there are no ratings, that is, using Mahout
terms we have boolean preferences.

Mahout booleanpreferences recommender returns a value for each unconsumed
item so that we can rank them. However, we cannot compare these values with
the items the user already consumed. Already consumed items have no
predicted value.

I hope I could explain the problem.
Thanks in advance.

Re: Recommending already consumed items

[Manuel Blechschmidt <Ma...@gmx.de>]

Hi Mario,
you can just provide 2 different group of recommendations:

1. Your personal favorites
2. You might also like.

This is how I solved this problem for another project.

/Manuel

On 02.12.2013, at 21:22, Mario Levitin wrote:

> Hi all,
> In some recommender applications the system might recommend already
> consumed items. For example, a hotel recommendation site might recommend
> hotel A to a user who already stayed at hotel A before.
> 
> In order to recommend already consumed items we have to rank all of the
> items (consumed and unconsumed ones)
> 
> My problem is how can we rank all the items? Let me explain.
> 
> If ratings are available I think the problem is easily solved. We can rank
> all the items because we have both the ratings given by the user and the
> predicted ratings.
> 
> However, what should we do if there are no ratings, that is, using Mahout
> terms we have boolean preferences.
> 
> Mahout booleanpreferences recommender returns a value for each unconsumed
> item so that we can rank them. However, we cannot compare these values with
> the items the user already consumed. Already consumed items have no
> predicted value.
> 
> I hope I could explain the problem.
> Thanks in advance.

-- 
Manuel Blechschmidt
M.Sc. IT Systems Engineering
Dortustr. 57
14467 Potsdam
Mobil: 0173/6322621
Twitter: http://twitter.com/Manuel_B