[packagekit] [GSOC] Recommender Systems

[Lucas Moura]

Hello,

Currently, AppRecommender uses two main strategies for content based
recommendation. A term frequency-inverse document frequency (TFIDF) method
to filter the most significant terms of the user installed packages and
query Xapian with this list. The information we apply the TFIDF one is the
package description, section and its debtags, which are the one we believe
has the highest amount of information. The second strategy is to perform a
 query expansion, where the installed packages are passed to Xapian, which
performs a query expansion on them to search for new packages.

On my bachelor thesis, a friend and I are trying to add context information
to the selected packages, by looking at the most recent used packages and
using a machine learning approach to better filter the recommended packages
based on this contextual information. But this approach is not stable yet.
This strategy is based on the TFIDF approach.

Best regards,
Lucas Moura

On Tue, Apr 26, 2016 at 4:57 PM, Richard Hughes <hughsient at gmail.com> wrote:

> On 26 April 2016 at 20:42, Lucas Moura <lucas.moura128 at gmail.com> wrote:
> > format, specially on the control file of a debian package, since it uses
> > info such as the package section, description and debtags.
>
> What algorithm were you thinking of using? We try to do something
> smart in gnome-software but we only use the category information to
> try to find similar software.
>
> Richard.
> _______________________________________________
> PackageKit mailing list
> PackageKit at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/packagekit
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/packagekit/attachments/20160426/9ca3f0bb/attachment.html>