predictionio-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Pat Ferrel <...@occamsmachete.com>
Subject Re: Decrease score of user viewed item instead of excluding
Date Sun, 25 Dec 2016 18:18:32 GMT
No. I think this might make sense but my first question is are you using multiple indicators?
If you have a lot of users that only get recommendations they already know about, you are
not getting a lot of value from the recommender—adding more data might help.

If you’ve done the above, consider that you already have some record of things the user
has viewed (maybe in an external DB). It is often more practical to blacklist all of these
and then mix-on some fixed number after you get recommendations. You could also create 2 engines,
one with a blacklist and the other without—the difference in recs would be the blacklisted
items so even if you don’t have a record of them somewhere else (in a db) you can re-create
it. Then mix them as you wish.

The final option would be to modify the code. The UR takes a PIO-type query then forms an
elasticsearch query from that along with data in the EventServer (including history of views).
It would be pretty easy to give a “boost” of 0 < n < 1 to lower viewed items in
the recommended items ranking. The UR internally puts them into a “should_not” clause,
all you would do is add them to the “should” clause and boost them by a fraction so the
score would be lowered. 

If you end up doing this, it could be donated back to the project if you also make it general
by adding something to the config in engine.json to specify the boost value and turn the feature
on. If you want help start a PR here: https://github.com/actionml/universal-recommender <https://github.com/actionml/universal-recommender>


On Dec 23, 2016, at 10:58 PM, Lin Amy <linamy85@gmail.com> wrote:

Hello everyone,

I am using universal recommendation template, and since sometimes the number of item isn't
that much, so user may have viewed all of items in given query, which leads to no results
for recommendation.
So my question is that, is there any advise on how to decrease the score of user viewed items
instead of sending them into blacklist?

Thank you for any advises!!

Best regards,
Amy


Mime
View raw message