lucene-solr-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Á_____o <>
Subject Combining Solr score with customized user ratings for a document
Date Wed, 13 Feb 2013 12:20:30 GMT

I am working on a proyect where we want to recommend our users products
based on their previous 'likes', purchases and so on (typical stuff of a
recommender system), while we want to let them browse freely the catalogue
by search queries, making use of facets, more-like-this and so on (typical
stuff of a Solr index).

After reading here and there, I have reached the conclusion that's it's
better to keep Solr Index apart from the database. Solr is for products
(which can be reindexed from the DB as a nightly batch) while the DB is for
everything else, including -the products and- user profiles. 

So, given an user and a particular search (which can be as simple as "q=*"),
on one hand we have Solr results (i.e. docs + scores) for the query, while
on the other we have user predicted ratings (i.e. recommender scores) coming
from the DB (though they could be cached elsewhere) for each of the products
returned by Solr.

And what I want is clear -to state-: combine both scores (e.g. by a simple
product) so the user receives a sorted list of relevant products biased by
his/her preferences.

I have been googleing for the last days without finding which is the best
way to achieve this.

I think it's not a matter of boosting, or at least I can't see which
boosting method could be useful as the boost should be user-based. I think
that I need to extend -somewhere- Solr so I can alter the result scores by
providing the user ID and connecting to the DB at query time, doing the
necessary maths and returning the final score in a -quite- transparent way
for the Web app.

A less elegant solution could be letting Solr do its work as usual, and then
navigate through the XML modifying the scores and reordering the whole list
of products (or maybe just the first N results) by the new combined score.

What do you think?
A big THANKS in advance


View this message in context:
Sent from the Solr - User mailing list archive at

View raw message