taste-users Mailing List for Taste

SourceForge Headquarters 225 Broadway Suite 1600 San Diego, CA 92101 +1 (858) 454-5900

Hi -

I was wondering if Taste could provide real time recommendations to
users? Real time meaning, if we have hundreds of thousands of ratings, a
new users goes and rates some items, can we provide them with valuable
feedback immediately without calling refresh on the recommender? we are
currently using a flat file implementation.

cheers
paul

Yes, well, for at least one implementation this is realistic.

There are two methods on Recommender that let you update preferences:
setPreference() and removePreference(). Why here, instead of
DataModel? in fact DataModel has similar methods. The idea is that
Recommender might be able to incorporate the update without a full
refresh if it is informed. The existing implementations pass on the
update to the DataModel too.

It turns out that really SlopeOneRecommender is the best at
incorporating updates on the fly. It should be very fast to update and
does not need a refresh. The correlation-based recommenders have to
clear all their cached correlations for the user who just got updated,
which isn't a full refresh, but is more significant.

Now you are using a FileDataModel, which isn't updateable, and is
going throw an exception if it receives an update. For the moment, you
could comment that out and just make these methods do nothing. I'll
need a little bit to think about how to handle this -- on the one hand
FileDataModel shouldn't act like it is updateable, but, maybe in this
situation it should be possible to let the in-memory Recommender
update independently of the DataModel. Hopefully your updates are
getting down into the underlying file at some point anyway.

Sean

On 10/23/07, pauls <tas...@qu...> wrote:
> Hi -
>
> I was wondering if Taste could provide real time recommendations to
> users? Real time meaning, if we have hundreds of thousands of ratings, a
> new users goes and rates some items, can we provide them with valuable
> feedback immediately without calling refresh on the recommender? we are
> currently using a flat file implementation.
>
> cheers
> paul
>
> -------------------------------------------------------------------------
> This SF.net email is sponsored by: Splunk Inc.
> Still grepping through log files to find problems?  Stop.
> Now Search log events and configuration files using AJAX and a browser.
> Download your FREE copy of Splunk now >> http://get.splunk.com/
> _______________________________________________
> Taste-users mailing list
> Tas...@li...
> https://lists.sourceforge.net/lists/listinfo/taste-users
>

Hi -

I was wondering if Taste could provide real time recommendations to 
users? Real time meaning, if we have hundreds of thousands of ratings, a 
new users goes and rates some items, can we provide them with valuable 
feedback immediately without calling refresh on the recommender? we are 
currently using a flat file implementation.

cheers
paul

There is already an implementation of Preference that has time
information -- DetailedPreference. None of the code uses it for two
reasons -- I'm very memory-conscious and adding another long to each
of millions of preference objects could be expensive. Also applying
time information to these algorithms is somewhat application-specific.

You can use this object in your own implementations though for
convenience. I'd also be open to creating some utility classes related
to time information. For example: a Rescorer which discounts the
preference based on age.

Sean

On 5/21/07, Cristiano Longo <cri...@ya...> wrote:
> In order to evaluate a collaborative filtering
> algorithm, I think should be nice to add temporal
> information to Preferences. Should be enough to define
> an ordering to the preferences in a data set(i.e.
> Preference implements Comparable).
>
>
>
>
>
>
>
>
> ___________________________________
> L'email della prossima generazione? Puoi averla con la nuova Yahoo! Mail:
> http://it.docs.yahoo.com/nowyoucan.html
>
> -------------------------------------------------------------------------
> This SF.net email is sponsored by DB2 Express
> Download DB2 Express C - the FREE version of DB2 express and take
> control of your XML. No limits. Just data. Click to get it now.
> http://sourceforge.net/powerbar/db2/
> _______________________________________________
> Taste-users mailing list
> Tas...@li...
> https://lists.sourceforge.net/lists/listinfo/taste-users
>

In order to evaluate a collaborative filtering
algorithm, I think should be nice to add temporal
information to Preferences. Should be enough to define
an ordering to the preferences in a data set(i.e.
Preference implements Comparable).

___________________________________ 
L'email della prossima generazione? Puoi averla con la nuova Yahoo! Mail: 
http://it.docs.yahoo.com/nowyoucan.html

Hi Cristiano can you post the code you used to run this, and your
data? I imagine there is not enough data here and the training set or
test set is empty for each user, or the data is split across the two
in a way that correlations can't be computed in the training set. You
can also verify this by stepping through with a debugger. Or just make
up a lot more data!

On 5/10/07, Cristiano Longo <cri...@ya...> wrote:
> Morning all,
> using AverageAbsoluteDifferenceRecommenderEvaluator
> and RMSRecommenderEvaluator I receive Not A Number as
> result.
>
> DataModel = GenericDataModel
> Correlation =  PearsonCorrelation
> Neighborhoos = NearestNUserNeighborhood with one
> neighbor
>
> Three items and four users that voted at least two
> items.
>
> Thanks in advance,
> Cristiano Longo
>
>
>       ___________________________________
> L'email della prossima generazione? Puoi averla con la nuova Yahoo! Mail: http://it.docs.yahoo.com/nowyoucan.html
>
> -------------------------------------------------------------------------
> This SF.net email is sponsored by DB2 Express
> Download DB2 Express C - the FREE version of DB2 express and take
> control of your XML. No limits. Just data. Click to get it now.
> http://sourceforge.net/powerbar/db2/
> _______________________________________________
> Taste-users mailing list
> Tas...@li...
> https://lists.sourceforge.net/lists/listinfo/taste-users
>

Morning all, 
using AverageAbsoluteDifferenceRecommenderEvaluator
and RMSRecommenderEvaluator I receive Not A Number as
result.

DataModel = GenericDataModel
Correlation =  PearsonCorrelation
Neighborhoos = NearestNUserNeighborhood with one
neighbor

Three items and four users that voted at least two
items.

Thanks in advance,
Cristiano Longo

      ___________________________________ 
L'email della prossima generazione? Puoi averla con la nuova Yahoo! Mail: http://it.docs.yahoo.com/nowyoucan.html

SoftCache is the SoftReference-based cache object that's used in
several places... and I think I'm going to scrap it. It introduces
noticeable overhead and doesn't help that much -- it only adds value
when you run out of memory, and when you do, well, it's not the best
thing to just flush all caches. I am replacing with a hot-rodded Map
implementation called FastMap that instead has a maximum size and a
fast, if simple, eviction policy.

The theory is that DataModel itself does not cache. It's always
delivering the most recent data from backing store.

Higher layers most certainly should cache. All the given
implementations do some sensible caching for you. While you may want
to add caching to your own implementations, you shouldn't have to
worry about doing that for given implementations.

The caches themselves don't copy anything. I would not rely on any
particular connection between your data and the objects -- this isn't
like Hibernate where setting values on the data objects affects the
backing store. None of the implementations I made work like that -- in
fact they try to be immutable as much as possible.

I'm not sure if I answered your question there.

BDB = Berkeley DB? I'd follow the patterns you see in the
JDBCDataModel implementations for such an implementation. I

On 4/30/07, karl wettin <kar...@gm...> wrote:
> What is the best way to deliver model data?
>
> I know there is a soft- or weak referenced thingy in there somewhere,
> but what does it keep track of? The instances of Preference, User and
> Item I created? Or does the cache actually copy the values from my
> instances and then dispose of them?
>
> For instance, should I leave context to my data source in a User
> implementation for Perference iteration, or should I create a set
> with all the Preference:s when I create the User?
>
> I'm attempting to implement a BDB data model.
>
> --
> karl
>
> -------------------------------------------------------------------------
> This SF.net email is sponsored by DB2 Express
> Download DB2 Express C - the FREE version of DB2 express and take
> control of your XML. No limits. Just data. Click to get it now.
> http://sourceforge.net/powerbar/db2/
> _______________________________________________
> Taste-users mailing list
> Tas...@li...
> https://lists.sourceforge.net/lists/listinfo/taste-users
>

What is the best way to deliver model data?

I know there is a soft- or weak referenced thingy in there somewhere,  
but what does it keep track of? The instances of Preference, User and  
Item I created? Or does the cache actually copy the values from my  
instances and then dispose of them?

For instance, should I leave context to my data source in a User  
implementation for Perference iteration, or should I create a set  
with all the Preference:s when I create the User?

I'm attempting to implement a BDB data model.

-- 
karl

On Thu, 2006-05-18 at 08:55 +0900, Sean Owen wrote:
> I see, it seems like you really want to cluster your users, then
> determine the most-preferred documents for each cluster. The
> UserNeighborhood implementations don't quite help you do that --
> they'll tell you which users are close to a given user, but won't show
> you the best set of disjoint neighborhoods that cover all users.

Aha. 

A social network is a neighborhood in this sense? Maybe I could use this
to estimate how much I trust someone that is not wired in a social
network.

> The UserCorrelation implementations would surely help you implement
> any standard clustering algorithm. You could copy/re-use code from
> Taste that figures out an average preference for an Item across a
> group of Users. Then it would be relatively straightforward to
> implement a new Recommender -- for a User, figure out which cluster
> the User belongs to, then return top recommendations for that cluster.
> 
> Any interest in writing this? :) 

Yes. But I can't promise that I'll do it. So many projects. So little I
really know about CF. I would need help.

Do you read ASCII class diagrams? My project looks like this:

[Clzifd]<#>- {0..*} | {0..*} -[Clazzifier]- {0..*} --<#>[Clazz]
                    |
            [Clazzification]

For instance, Clazzified A has a Clazzification in clazzifier "horror"
of Clazz "movie generes". 

Documents returned from a free text search are Clazzified. The
clazzifications are the semantic description of these things.

Users could also be clazzified. A clazzification could then be
knowledge, interest, et.c.

So it would be nice to fit Taste in there. If it could build me a pseudo
social network, it would be cool.

> I can otherwise put it on the to-do list.

Would you actually spend time on this? That would be great. 

> 
> Yeah Taste says nothing about where you get user preferences -- there
> is no reason they have to be explicitly collected from the user, and
> in fact I think most effective CF systems will try to figure
> preference from much more than just the user's explicit feedback. All
> of your ideas sound good. There's no one right answer for it, since
> guessing a preference is very application-specific.

Maybe my question was not very well formed :) What I really asked was if
I need negative training for Taste to work well.

-- 
karl

> 
> Sean
> 
> 
> On 5/18/06, karl wettin <ka...@sn...> wrote:
> > Sean,
> >
> > it would be great to hear your thoughts on my plan. I want to boost
> > results from Apache Lucene using CF.
> >
> > It would be great if I could simply associate each of my documents
> > (items) with a number of neighborhoods. If the user is a member of any
> > of these, I boost the results accordingly. I would then recalculate the
> > CF-stuff when the load is low. Is this doable? What will I loose by
> > doing it that way?
> >
> > Also, I was thinking to not have the users explicitly having to tell me
> > their preferences. There are a bunch of things that is kind of an
> > indication of what they like. What stuff they keep searching for,
> > categories they upload to, things they download, if they upload the same
> > thing as someone else, if they comment uploads as "GREAT STUFF, THANK
> > YOU INDEED", et.c.
> >
> > I don't have very good indications on what they do not like.
> >
> > Do you think this is enough for good results?
> >
> >
> >
> >
> > -------------------------------------------------------
> > Using Tomcat but need to do more? Need to support web services, security?
> > Get stuff done quickly with pre-integrated technology to make your job easier
> > Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
> > http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642
> > _______________________________________________
> > Taste-users mailing list
> > Tas...@li...
> > https://lists.sourceforge.net/lists/listinfo/taste-users
> >
> 
> 
> -------------------------------------------------------
> Using Tomcat but need to do more? Need to support web services, security?
> Get stuff done quickly with pre-integrated technology to make your job easier
> Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
> http://sel.as-us.falkag.net/sel?cmd=lnk&kid0709&bid&3057&dat1642
> _______________________________________________
> Taste-users mailing list
> Tas...@li...
> https://lists.sourceforge.net/lists/listinfo/taste-users
> 

I see, it seems like you really want to cluster your users, then
determine the most-preferred documents for each cluster. The
UserNeighborhood implementations don't quite help you do that --
they'll tell you which users are close to a given user, but won't show
you the best set of disjoint neighborhoods that cover all users.

The UserCorrelation implementations would surely help you implement
any standard clustering algorithm. You could copy/re-use code from
Taste that figures out an average preference for an Item across a
group of Users. Then it would be relatively straightforward to
implement a new Recommender -- for a User, figure out which cluster
the User belongs to, then return top recommendations for that cluster.

Any interest in writing this? :) I can otherwise put it on the to-do list.

Yeah Taste says nothing about where you get user preferences -- there
is no reason they have to be explicitly collected from the user, and
in fact I think most effective CF systems will try to figure
preference from much more than just the user's explicit feedback. All
of your ideas sound good. There's no one right answer for it, since
guessing a preference is very application-specific.

Sean

On 5/18/06, karl wettin <ka...@sn...> wrote:
> Sean,
>
> it would be great to hear your thoughts on my plan. I want to boost
> results from Apache Lucene using CF.
>
> It would be great if I could simply associate each of my documents
> (items) with a number of neighborhoods. If the user is a member of any
> of these, I boost the results accordingly. I would then recalculate the
> CF-stuff when the load is low. Is this doable? What will I loose by
> doing it that way?
>
> Also, I was thinking to not have the users explicitly having to tell me
> their preferences. There are a bunch of things that is kind of an
> indication of what they like. What stuff they keep searching for,
> categories they upload to, things they download, if they upload the same
> thing as someone else, if they comment uploads as "GREAT STUFF, THANK
> YOU INDEED", et.c.
>
> I don't have very good indications on what they do not like.
>
> Do you think this is enough for good results?
>
>
>
>
> -------------------------------------------------------
> Using Tomcat but need to do more? Need to support web services, security?
> Get stuff done quickly with pre-integrated technology to make your job ea=
sier
> Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronim=
o
> http://sel.as-us.falkag.net/sel?cmd=3Dlnk&kid=3D120709&bid=3D263057&dat=
=3D121642
> _______________________________________________
> Taste-users mailing list
> Tas...@li...
> https://lists.sourceforge.net/lists/listinfo/taste-users
>

Sean,

it would be great to hear your thoughts on my plan. I want to boost
results from Apache Lucene using CF. 

It would be great if I could simply associate each of my documents
(items) with a number of neighborhoods. If the user is a member of any
of these, I boost the results accordingly. I would then recalculate the
CF-stuff when the load is low. Is this doable? What will I loose by
doing it that way?

Also, I was thinking to not have the users explicitly having to tell me
their preferences. There are a bunch of things that is kind of an
indication of what they like. What stuff they keep searching for,
categories they upload to, things they download, if they upload the same
thing as someone else, if they comment uploads as "GREAT STUFF, THANK
YOU INDEED", et.c.

I don't have very good indications on what they do not like.

Do you think this is enough for good results?

On Sun, 2006-05-14 at 00:24 -0500, Sean Owen wrote:
> On 5/13/06, karl wettin <ka...@sn...> wrote:
> > public interface User extends Comparable<User> {
> >         /**
> >          * @return unique user ID
> >          */
> >         Object getID();
> >
> >
> > Whats the deal with this ID and what is wrong with User.equals() and
> > Item.equals()?

> You're asking why you would need an ID object, instead of just
> comparing User / Item implementation objects directly?

Exactly. I don't have an ID. The instance is my identity.

Thinking about it I could just return "this" as ID.

You're asking why you would need an ID object, instead of just
comparing User / Item implementation objects directly?

GenericUser / GenericItem implements equals() and hashCode() based on
ID, so yeah you could treat them this way if you wanted. That may not
be true of other implementations. I could have either written that you
must implement equals()/hashCode()/compareTo() sensibly, or else force
you to provide an "ID" object like this. I guess I thought the latter
made most sense since just about every application has an idea of user
ID and item ID.

But that's about as far as I thought about it. Is this problematic in
your application?

Sean

On 5/13/06, karl wettin <ka...@sn...> wrote:
> public interface User extends Comparable<User> {
>         /**
>          * @return unique user ID
>          */
>         Object getID();
>
>
> Whats the deal with this ID and what is wrong with User.equals() and
> Item.equals()?
>
>
>
> -------------------------------------------------------
> Using Tomcat but need to do more? Need to support web services, security?
> Get stuff done quickly with pre-integrated technology to make your job ea=
sier
> Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronim=
o
> http://sel.as-us.falkag.net/sel?cmd=3Dlnk&kid=3D120709&bid=3D263057&dat=
=3D121642
> _______________________________________________
> Taste-users mailing list
> Tas...@li...
> https://lists.sourceforge.net/lists/listinfo/taste-users
>

public interface User extends Comparable<User> {
	/**
	 * @return unique user ID
	 */
	Object getID();

Whats the deal with this ID and what is wrong with User.equals() and
Item.equals()?

13 nov 2005 kl. 20.33 skrev Sean Owen:

> This is most definitely a good idea -- this is exactly the kind of
> thing that CF should be used for.

I started writing this crazy software where CF is only one of many  
parts. Right now it looks like an RSS-reader, but it is supposed to  
become so much more.

You can check it out at <http://82.93.70.124:11235/>. Usually up all  
the time, but I do restart now and then.

Most feeds are in Swedish, but it is easy to import your own OPML.  
I'm now in need of help creating a test database with ratings. If you  
decide to create an account, please email me so that I know. Don't  
hesitate to contact me when you don't understand, find bugs, want  
features et.c. I'll be coding all weekend.

This is what I wrote about the software. I'm not even close to  
implement all the features described.

myWhat?
This software gathers contents and meta data about the contents.
Content can be anything. Local media files, RSS feeded articles,  
forum posts, mail, et.c.
Meta data can be discussions, ratings, reviews, subject  
classifications and semantic classifications
The system try to learn what you are interested in right now by using  
vector classifications and collaborative filtering (CF). It is (wich  
most CF implementations is not) senisble to how your interest has  
changed over time. You may conider a movie be bad today even if you  
you really liked twenty years ago.
By comparing interest curvs between profiles over time, myEchelon  
will find trends that can lead it to what you will be likely to find  
be interesting in the future.
Subject is also an important factor. Two people can share views  
within music but not within movies at the same time.
It is possible to semantically relatate any data. A post can be the  
reply to the other, but it could also be the solution of a problem.  
If in your search for a solution find the same a question without the  
answer, that post could be semantically related as the same question  
to a post with a solution related to it. And thus you find the  
solution to any resolved problem without the hassle of besserwissers  
telling you to RTFM.

It does require karma to work smooth. You must trust other people  
with potentially sensitive data about yourself. There is no central  
storage of data for calcualtions. It is done on your computer with  
the information feeded by you and the people how trust you. Everybody  
will run their own server that will be filled with content by people  
they trust, and only with content their friends sincerely thought  
they were interested in.
If you read something new and interesting, you recommend it to your  
friends. Unless your friends has the permission to, the rating will  
not be passed on. In order for the content to be populated, your  
friend have to create an own rating it and pass that rating on to its  
friends.
So these trust chains also have the side effect to halt evil  
commercialism and propaganda.

All these things are pretty cool by themselfs. I hope for awesome  
query possibillities when one combines them.

Find an answer to my question within usenet posts by people like me,  
order by how much I trust them to have had the knowledge in the  
subject at the time of the post
Play some jazz everybody at the party likes and order by the tunes  
nobody heard before.
What drink should I make of the current supply at the party at this  
time when I'm this buzzed?
about

Runs on a Java 1.5 virtual machine.
All servers (RDBMS, HTTP, et.c.) are built in, running within the  
same JVM for minimum use of clock ticks.
Hibernate persistence transparently intercepted by Lucene for text  
searches.
Classifier4J does the vector (subject) classification,
and collaborative filtering will probably be done using Taste.
The very nice Jakarta sandbox projects Feedparser and Thread pool are  
also used.

> There are three parts to building a
> system like this:
>
> 1. Create an app/plugin that can send usage information (which RSS
> feeds you subscribe to, how often you read them, etc.) to a server and
> later download recommendations.
>
> 2. Translate usage information into some idea of preference.
>
> 3. Compute recommendations.
>
> Taste addresses #3 directly.
>
> #2 can be solved if you ask the user to explicitly rate each feed (or
> even article), but, in my opinion CF-based applications need to get
> away from this. It's a burden on the user. The interesting question is
> how to guess a preference implicitly based on usage info and I haven't
> seen much research on that. If I know you have subscribed to a feed
> and read it once a month, what do I figure your opinion of it is?
> versus a feed you read daily? You probably like the latter more, but
> how much more? how would I know you hate a feed?
>
> #1 is the part I'm not so familiar with. I would rather not write Yet
> Another RSS Reader but would like to add such capability to an
> existing reader somehow. You have experience with RSS-owl? anyone else
> have ideas on this point?
>
> Sean
>
>
> On 11/13/05, karl wettin <ka...@sn...> wrote:
>> Sean,
>>
>> it would be cool with an channel reader featuring collaborative
>> filtering. Preferably peer-to-peer. RSS, usenet, email, et.c.
>>
>> Any thoughts on that?
>>
>> I might have to patch up RSS-owl.
>>
>> --
>> karl
>>
>>
>> -------------------------------------------------------
>> SF.Net email is sponsored by:
>> Tame your development challenges with Apache's Geronimo App  
>> Server. Download
>> it for free - -and be entered to win a 42" plasma tv or your very own
>> Sony(tm)PSP.  Click here to play: http://sourceforge.net/geronimo.php
>> _______________________________________________
>> Taste-users mailing list
>> Tas...@li...
>> https://lists.sourceforge.net/lists/listinfo/taste-users
>>
>
>
> -------------------------------------------------------
> SF.Net email is sponsored by:
> Tame your development challenges with Apache's Geronimo App Server.  
> Download
> it for free - -and be entered to win a 42" plasma tv or your very own
> Sony(tm)PSP.  Click here to play: http://sourceforge.net/geronimo.php
> _______________________________________________
> Taste-users mailing list
> Tas...@li...
> https://lists.sourceforge.net/lists/listinfo/taste-users
>

I just put out another release, collecting a number of small fixes and
refinements. Victor I included your API change, so that Recommender
now returns List<RecommendedItem>. Enjoy!

Sean

Yeah that's a very good idea, and easy to implement. I'll have to
think a bit about how to change the public API (and I'm open to
suggestions). Give me a few days and I'll send you an updated JAR.

Sean

On 11/22/05, Victor Bayon <vb...@gm...> wrote:
> Hi Sean,
>  thanks for your reply. I am afraid I am not a "hard core" user of your t=
ool
> right now and I am not sure if I could give you useful feedback at this
> stage.
>
>  I am wondering if you can get a  "percentage" recomendation in addition =
to
> the list that you get, lts
>
>  I get as an output now
>  Item[78394578943]
>  Item[78394573443]
>  .....
>  ......
>
>
>  Item[78394578943] - 80%
>  Item[78394573443] - 60%
>  .........
>  .......
>
>
>  this could be useful if you are doing a user interface to represent
> visually the recomendation.
>
>
>  thanks and regards
>
>  /Victor

This is most definitely a good idea -- this is exactly the kind of
thing that CF should be used for. There are three parts to building a
system like this:

1. Create an app/plugin that can send usage information (which RSS
feeds you subscribe to, how often you read them, etc.) to a server and
later download recommendations.

2. Translate usage information into some idea of preference.

3. Compute recommendations.

Taste addresses #3 directly.

#2 can be solved if you ask the user to explicitly rate each feed (or
even article), but, in my opinion CF-based applications need to get
away from this. It's a burden on the user. The interesting question is
how to guess a preference implicitly based on usage info and I haven't
seen much research on that. If I know you have subscribed to a feed
and read it once a month, what do I figure your opinion of it is?
versus a feed you read daily? You probably like the latter more, but
how much more? how would I know you hate a feed?

#1 is the part I'm not so familiar with. I would rather not write Yet
Another RSS Reader but would like to add such capability to an
existing reader somehow. You have experience with RSS-owl? anyone else
have ideas on this point?

Sean

On 11/13/05, karl wettin <ka...@sn...> wrote:
> Sean,
>
> it would be cool with an channel reader featuring collaborative
> filtering. Preferably peer-to-peer. RSS, usenet, email, et.c.
>
> Any thoughts on that?
>
> I might have to patch up RSS-owl.
>
> --
> karl
>
>
> -------------------------------------------------------
> SF.Net email is sponsored by:
> Tame your development challenges with Apache's Geronimo App Server. Downl=
oad
> it for free - -and be entered to win a 42" plasma tv or your very own
> Sony(tm)PSP.  Click here to play: http://sourceforge.net/geronimo.php
> _______________________________________________
> Taste-users mailing list
> Tas...@li...
> https://lists.sourceforge.net/lists/listinfo/taste-users
>

Sean,

it would be cool with an channel reader featuring collaborative  
filtering. Preferably peer-to-peer. RSS, usenet, email, et.c.

Any thoughts on that?

I might have to patch up RSS-owl.

-- 
karl

0.8 (2005-09-11)
- Added a mechanism to create and deploy a web interface to Taste
  on top of a servlet container like Tomcat
  - Supports web services via Axis
  - Supports simpler HTTP GETs for recommendations too
- Rewrote the EJB support to match the new HTTP web interface
- Minor fixes and slight performance tweaks across the board

Oops, I forgot to forward my reply to the list.

Also, I have just posted 0.7.5, which increases performance a bit, but
also lets you use pre-computed item similarity values with item-based
recommenders, which greatly increases performance. That's really the
right way to use item-based recommenders anyway.

Sean

On 9/3/05, Sean Owen <sr...@gm...> wrote:
> It's up! this weekend I am going to release another small revision to
> improve performance, because that remains the biggest issue. I have
> been using a -very- basic sample web app on Tomcat, using MovieLens
> data, to test. I can send that to you separately if it's useful,
> though I won't yet put it in the distribution. I do want a decent
> example application in the distribution later.
>=20
> Right now the code includes MySQLJDBCDataModel (attached), which
> implements DataModel on top of a database. In there you can see the
> kinds of queries it constructs -- it's pretty simple, and could
> probably be used with any standard SQL database. It's just looking for
> a table with three columns -- user ID, item ID, preference value.
>=20
> If this doesn't work for your DB, let me know and we'll work out a way
> to refactor it so that it does work.
>=20
> Sean
>=20
>=20
> On 9/3/05, Karl Wettin <ka...@sn...> wrote:
> > is the list up now?
> >
> > i havent even looked at taste more than the webpage. are there any
> > example apps? again, i would like to use an oodbms as data source:
> > how complex queries is required in order for a datasource to work? is
> > it enough to basiaclly return a data iterator or so?
> >
> > --
> >
> >    karl
> >
> >
> > -------------------------------------------------------
> > SF.Net email is Sponsored by the Better Software Conference & EXPO
> > September 19-22, 2005 * San Francisco, CA * Development Lifecycle Pract=
ices
> > Agile & Plan-Driven Development * Managing Projects & Teams * Testing &=
 QA
> > Security * Process Improvement & Measurement * http://www.sqe.com/bsce5=
sf
> > _______________________________________________
> > Taste-users mailing list
> > Tas...@li...
> > https://lists.sourceforge.net/lists/listinfo/taste-users
> >
>=20
>=20
>

is the list up now?

i havent even looked at taste more than the webpage. are there any  
example apps? again, i would like to use an oodbms as data source:  
how complex queries is required in order for a datasource to work? is  
it enough to basiaclly return a data iterator or so?

-- 

   karl

This is a test.

2005	Jan	Feb	Mar	Apr	May	Jun	Jul	Aug (1)	Sep (3)	Oct	Nov (4)	Dec (1)
2006	Jan	Feb	Mar	Apr	May (6)	Jun	Jul	Aug	Sep	Oct	Nov	Dec
2007	Jan	Feb	Mar	Apr (2)	May (4)	Jun	Jul	Aug	Sep	Oct (4)	Nov	Dec
2008	Jan	Feb	Mar	Apr	May	Jun	Jul	Aug	Sep	Oct (8)	Nov (40)	Dec

taste-users Mailing List for Taste

taste-users — Open discussion among Taste users