Taste

[Nishant Chandra <ni...@gmail.com>]

My query is regarding taste APIs.

If I have binary dataset (i.e. preference is 0 or 1), are there models
other than TanimotoCoefficient?

Also, with the current implementation, is it possible to precompute
TanimotoCoefficient similarities and feed it into
ItemBasedRecommender? How?

Thanks,
Nishant

Re: Taste

[Sean Owen <sr...@gmail.com>]

What else would you like to know? I think the javadoc says everything
to say about the implementation. If you want to know more about the
calculation, try wikipedia.

You can do what you say, but you'd have to code this yourself. It's
not hard at all.

I'd still maybe suggest you let the implementation intelligently store
similarities by caching rather than precomputing -- maybe more
effective -- but up to you.

On Mon, Jan 25, 2010 at 5:39 PM, Nishant Chandra
<ni...@gmail.com> wrote:
> Is there any documentation on LogLikelihoodSimilarity other than what
> is mentioned in the javadoc? I am unable to download the paper.
>
> The way I was thinking was to precompute the similarities and dump it
> in a file. And then let the GenericItemSimilarity read from this file.
> This was I can seperate the calculation and refresh recommender once I
> have a new set of calculated similarities. Any suggestions?

Re: Taste

[Ted Dunning <te...@gmail.com>]

Try here: http://tdunning.blogspot.com/2008/03/surprise-and-coincidence.html

On Mon, Jan 25, 2010 at 9:39 AM, Nishant Chandra
<ni...@gmail.com>wrote:

> Is there any documentation on LogLikelihoodSimilarity other than what
> is mentioned in the javadoc? I am unable to download the paper.
>



-- 
Ted Dunning, CTO
DeepDyve

Re: Taste

[Nishant Chandra <ni...@gmail.com>]

Is there any documentation on LogLikelihoodSimilarity other than what
is mentioned in the javadoc? I am unable to download the paper.

The way I was thinking was to precompute the similarities and dump it
in a file. And then let the GenericItemSimilarity read from this file.
This was I can seperate the calculation and refresh recommender once I
have a new set of calculated similarities. Any suggestions?

Thanks,
Nishant


On Mon, Jan 25, 2010 at 10:34 PM, Sean Owen <sr...@gmail.com> wrote:
> Yes, try LogLikelihoodSimilarity too.
>
> You can feed this similarity to GenericItemSimilarity to have it
> precompute and store all possible similarities, yes:
> http://lucene.apache.org/mahout/javadoc/core/org/apache/mahout/cf/taste/impl/similarity/GenericItemSimilarity.html#GenericItemSimilarity(org.apache.mahout.cf.taste.similarity.ItemSimilarity,
> org.apache.mahout.cf.taste.model.DataModel)
>
> You can also just use CachingItemSimilarity around it to cache, rather
> than precompute, which might be smarter.
>
> On Mon, Jan 25, 2010 at 3:35 PM, Nishant Chandra
> <ni...@gmail.com> wrote:
>> My query is regarding taste APIs.
>>
>> If I have binary dataset (i.e. preference is 0 or 1), are there models
>> other than TanimotoCoefficient?
>>
>> Also, with the current implementation, is it possible to precompute
>> TanimotoCoefficient similarities and feed it into
>> ItemBasedRecommender? How?
>>
>> Thanks,
>> Nishant
>>
>



-- 
Nishant Chandra
Hyderabad, India
Cell : +91 9949828480

Re: Taste

[Sean Owen <sr...@gmail.com>]

Yes, try LogLikelihoodSimilarity too.

You can feed this similarity to GenericItemSimilarity to have it
precompute and store all possible similarities, yes:
http://lucene.apache.org/mahout/javadoc/core/org/apache/mahout/cf/taste/impl/similarity/GenericItemSimilarity.html#GenericItemSimilarity(org.apache.mahout.cf.taste.similarity.ItemSimilarity,
org.apache.mahout.cf.taste.model.DataModel)

You can also just use CachingItemSimilarity around it to cache, rather
than precompute, which might be smarter.

On Mon, Jan 25, 2010 at 3:35 PM, Nishant Chandra
<ni...@gmail.com> wrote:
> My query is regarding taste APIs.
>
> If I have binary dataset (i.e. preference is 0 or 1), are there models
> other than TanimotoCoefficient?
>
> Also, with the current implementation, is it possible to precompute
> TanimotoCoefficient similarities and feed it into
> ItemBasedRecommender? How?
>
> Thanks,
> Nishant
>