Hello,

Which algorithm do you use for recommendations? Is there a performance comparison between the affinity module and the content recommendation module that uses the Slope One Algorithm for recommendations?

Thanks.

Comments

sikjoy’s picture

sikjoy commented

Hi jmseigneur:

I have not been able to successfully install the CRE module, so I can't make a fair comparison.

If votingapi is set up for percent:
The algorithm calculates the sum of the squares of the difference between vote vectors for each user.

If votingapi is set up for points:
The algorithm counts common votes.

The module uses cached results once a calculation has been performed. There are several conditions that must be met for a calculation to be performed:

1) The user must log in
2) they must have cast a configurable number of votes
3) the age of the cache must be older than a configurable amount of time
4) they must have cast a new vote since the last calculation
5) if the throttle module is enabled, the site must not be throttled

Furthermore, calculations are only done against the subset of users who have cast a minimum number of votes.

All this is done to mitigate the server load.

I have successfully tested it on a production server that gets ~10,000 page views per month. If you are able to benchmark this module, or run it on a more heavily trafficked site, please post the results here.

Cheers,

sicjoy

sikjoy’s picture

sikjoy commented
Status: Active » Closed (fixed)