mahout-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Mark <static.void....@gmail.com>
Subject Re: Boolean recommendations
Date Sun, 06 Mar 2011 23:56:20 GMT
Thanks for all the suggestions. Ill try some of these different 
approaches to see if any of them will work.

On 3/6/11 3:16 PM, Sean Owen wrote:
> That sounds slow indeed.
> CachingItemSimilarity around that LogLikelihoodSimilarity ought to
> help a fair bit.
> I'd also look at injecting a different CandidateItemsStrategy.
> PreferredItemsNeighborhoodCandidateItemsStrategy is the default, but
> SamplingCandidateItemsStrategy might work as well and it caps how much
> data it looks through to construct possible items to recommend, which
> will go straight to the issue of latency.
>
> On Sun, Mar 6, 2011 at 9:17 PM, Mark<static.void.dev@gmail.com>  wrote:
>> I have about 20 million boolean preferences and I am trying to suggest some
>> recommendations using a GenericBooleanPrefItemBasedRecommender with a
>> LogLikelihoodSimilarity. The results that are returned aren't terrible,
>> however it takes up to a minute for each recommendation.
>>
>> Unique users: ~ 600k
>> Unique items: ~ 15m
>>
>> Is there anything I can do to speed this up? Thanks
>>

Mime
View raw message