Home

AppRecommender is a project in development that aims to provide solutions for application recommendation at the GNU/Linux world. It was initially thought as a Debian package recommender, but considering the multi-distro effort in providing platform independent solutions, it should also follow this principle.

The main idea is to collect information from several data sources and provide package recommendations given the user list of installed packages in the system (previous choices).

From Debian, we can use the following data sources: apt-xapian-index, Debtags, Popcon, UDD etc. It might also be interesting to collect a user profile to refine the recommendations.

Some possible strategies:

• Content-based: recommend packages similar to the user installed packages, having package descriptions and debtags as data sources;
• Collaborative: recommend packages that similar users have installed, based on popcon data;
• Hybrid methods:
  • Refine recommendation using heuristics such as most popular packages, least buggy packages etc;
  • Refine recommendation by considering users profiles;
  • Collaboration through content;
  • Switch between basic recommenders, depending on the situation;
  • Combining results from different recommenders into an unique recommendation list.

Key points:

• How could we perform feature selection in this field?
• How packages dependencies interfere recommendations strategies?