I would like to request an improvement in the sourceforge.net project search algorithm. The current algorithm for "sort by relevance" lacks of enough intelligence to sort the results based on what most users expect to obtain.
For example, if you search for "computer vision" the results show projects which do match "computer vision" in the title or description, but which aren't really a good search results since they are not really active, or do not have any project content at all.
I can imagine a better algorithm that sort results based on some kind of weighted average of different factors such as: existence of project content, number weekly downloads, number of ratings, number of positive ratings, percentage of positive ratings, date of last update, update frequency, number of developers, existence of documentation, documentation size, etc...
it's seems that maybe the problem is not search algorithm itself but sort/filter capabilities?
maybe half of solution could be if user could construct his filter to represent info from forge db fields (Categories, License, User Ratings, etc..)
other half could be faceted search implementation?
the problem with projects with wrong Categories or other fields remains - solution for this could be feature 'suggest better ...(Category / decription /etc..)' alongside with "Report inappropriate content"