Nope, I tried that before and in my opinion the problem is the
CachingRecommender. I need a caching one because of the performance and
runtime. In fact, I must update or even clear the cache for the user who
updates his preferences. Information: I use a new CachingRecommender(new
SlopeOneRecommender(...), ...)
Additionally (I don't know if this matters), I use a MySQLDataModel but
no MySQLDiffStorage because it takes too long to build the diffs (after
3 hours it even hadn't finished!).
So, all in all, if you say the SlopeOneRecommender does it itself, my
problem is the CachingRecommender that caches all diffs if they are
called once until the cache gets cleared.
Am 24.02.2011 17:42, schrieb Sean Owen:
I see well if you are using SlopeOneRecommender, it already updates
the diffs correctly for you, when you call setPreference() for
example. If you do it that way it all works. There should be no need
to expose and update the diffs separately.
2011/2/24 Daniel Mühlbachler <[email protected]
<mailto:[email protected]>>
Ok but if I want to call removeItemPref I must know my preference
but to get that preference I need runtime. However, what happens
if I remove a preference and want to get the preference after that
again (so it should recalculate the user's preferences).
Another thing: if I want to call updateItemPref there is no way to
specify an user. Only itemID and a prefDelta. I understand why
there is no userID but what's prefDelta? And the biggest issue in
that case is the runtime because we have many queries at the same
time that must be processed as quickly as possible. In fact, if I
update all preferences is there a timeout for all other queries?
Ok, let's specify again what I have/need:
A website that interacts with my recommender and there will be
many many queries at the same time (huge site). In fact, a user
can rate movies and if a user does this we need a method that
rebuilds at least the user's recommendations. (All other
recommendations from other users that get affected by that change
are not important - only the user's ones because we are doing a
complete refesh every day.) For that, we can't have a high timeout
(e.g. refresh needs about 20-25 seconds). So the perfect way would
be only to update an user's preferences or the use of the
updateItemPref method (if this method can provide that things).
btw... thanks again for your help with my MySQL Connector and
Tomcat! :-)
Am 24.02.2011 17:15, schrieb Sean Owen:
It's just a count maintained across calls to the method which
tracks how
many entries have been stored in the data structures. This is
so it can stop
growing the data size once a limit has been reached.
It would not work to expose this method directly, as-is, no.
Instead I think you want to reflect your updates by calling
updateItemPref()
and/or removeItemPref() for all the changes that your user makes.
2011/2/24 Daniel Mühlbachler<[email protected]
<mailto:[email protected]>>
Hello,
I have a problem with the memory caching recommender. Due
to performance
issues I must choose a memory caching recommender but we
need to reload the
cache only for one user (I don't explain why).
After investigating the MemoryDiffStorage for the SlopeOne
recommender, I
found the method "private long processOneUser(long
averageCount, long
userID)" that processes only one user - right? In fact, my
idea is that I am
making this method public and add this method to the
WebService class but
what is this averageCount variable? Does it matter if I
choose a random
number or must that be a special one?
Moreover, where can I find the WebService class files and
how can I compile
them again to get a *.jws file so that I can add this
method to the
WebService?
Thanks in advance!
Best regards,
Daniel Mühlbachler
