|
SourceForge DBLens
Page (to download DBLens)
|
[October 10th, 2003] If you are looking for freely available
collaborative filtering recommendation engine, I recommend that you
consider the CFEngine instead. The CFEngine is my latest recommender
engine, this time written in Java. The CFEngine supports high-performance
operation, is much more stable, and much better supported. We are also
actively developing the CFEngine, while development on DBLens has ceased. You can download CFEngine from
http://eecs.oregonstate.edu/iis/
DBLens is a Oracle-based toolkit for performing collaborative
filtering. DBLens is a collection of PL/SQL code and accompanying
shell scripts that provides a flexible and efficient method for
performing collaborative filtering. The system was designed to perform
primarily batch offline computation. The parameters of the computation
are highly configurable.
Currently the system takes data in the form of numerical ratings of
the form:
<![CDATA[<userid, itemid, numeric rating>]]>
Given a set of ratings and
a request for a prediction
<![CDATA[<userid, itemid, ?>]]> the system will locate
users with similar interests, and use their ratings to predict the
unknown rating. There are also procedures for producing the top-n
highest predicted items for a given user, although these are not
thoroughly tested.
DBLens does all of it's computation inside Oracle using PL/SQL, and
thus an installation of Oracle on Unix is a necessity. DBLens has been
tested on Solaris, but should run on any Unix for which Oracle is
available.
Intended Audience
DBLens was designed as a tool for research into collaborative
filtering algorithms. Therefore the intended audience is people doing
research into collaborative filtering and information filtering. The
system was designed for batch usage, although DBLens can produce
predictions in real time - just not in volume.
License to Use
DBLens is released under a license very similar to that of BSD
Unix. Basically, you can use this software for just about anything, as
long as you maintain the copyright notices. See the licenses in the
source files for more information.
Who is responsible?
DBLens was primarily developed by
When initially developed, both were members of the GroupLens Research Group at the
University of Minnesota.
Downloading DBLens
You can download DBLens from SourceForge.
Open Source Development
DBLens has been "Open Sourced." If you would like to contribute code
or modifications, please contact Jon Herlocker. The CVS source
code tree is available through SourceForge.
Educational Licenses for Oracle
Don't despair! While commercial licenses cost big $$$, Oracle will
license its Oracle server product to educational institutions for a
reasonable sum of money (several hundred dollars).
Publicly Available Data
If you are looking to do some CF research, the next thing you will
probably need after getting DBLens is a dataset. GroupLens Research
has a set publicly
available, consisting of 100,000 ratings for 1682 movies by 943
users.
GroupLens Research
|