More search improvements

[ineiti]

Searching for eid showed our old eid-demo first, which is not ideal. Playing around with different parameters, I think the following would be the best:

sortProjects::sortProjects receives a filtered list of projects, then:

• a = mean(showcase_interest)
• b = searchRelevance // this is 0 if no search is performed
• c = 4 for 2026 Proposals, 3 if current C4DT activities, 2 if Lab activities, 1 if past C4DT activities, 0 else
• then normalize a, b, c to be between 0 and 1, inclusive
• relevance = a + 2 * b + 4 * c

This will:

• put all 2026 Proposals first
• except very relevant search results for projects with C4DT activities

[PascalinDe]

I think search should not make too many assumptions on what the person using it is actually looking for

• we have the carousel for selected projects that we want to highlight, so these projects are easily findable for anyone just browsing
• we already sort the projects by whether they are actively supported or not (if this sort order is not maintained in the search results, it's a bug), so the more "active" projects should already be sorted higher
• if we have projects that we consistently "sort down", I'd advocate for "phasing out" projects completely from the showcase, as the purpose of the showcase is to highlight what labs are doing now, and not be an archive (at least that's how I understand it)

if we tweak the search too much, I'm not sure if this will really helpful in the long run, as it'll become unintuitive where to look for in the results for what you're searching for - since we are currently the people using the showcase the most, we might be searching for old projects, etc.

[ineiti]

As we saw, if you define search as "just show the projects which include the search term anywhere" doesn't give useful results in all cases. So we need to make some assumptions. And here I propose something which came up while I was actually searching for different projects / professors / labs on the interface.

Your other point with regard to phasing out old projects is valid - currently I'm in "we have sooo many projects", but of course this doesn't make any sense if the projects are not active in any way. Then again, sometimes an old project all of a sudden gets popular again, and you never really know why. So I rely on the professors to tell me to remove projects from the list, which they rarely do :)

[PascalinDe]

As we saw, if you define search as "just show the projects which include the search term anywhere" doesn't give useful results in all cases. So we need to make some assumptions. And here I propose something which came up while I was actually searching for different projects / professors / labs on the interface.

it was not only just matching (although that part was really simplistic, I agreed with you), it should also sort by whether a project is being actively developped in the lab, and/or supported by C4DT - if that is not the case there is a bug somewhere

I guess I'm mostly worried about keeping the search results somewhat intuitive, i.e. including attributes in the calculation of the score that people can actually see - "this is further down because it's not active" - as opposed to "I don't understand why this project is further down, it looks the same to me as the other one"

Your other point with regard to phasing out old projects is valid - currently I'm in "we have sooo many projects", but of course this doesn't make any sense if the projects are not active in any way. Then again, sometimes an old project all of a sudden gets popular again, and you never really know why. So I rely on the professors to tell me to remove projects from the list, which they rarely do :)

we could keep the data in the project files, but not show them in the interface (e.g. by just not showing anything before a certain date or so) - if a topic becomes hot again, we can then just change this - I doubt our users will go off digging really deep down to find old projects

[ineiti]

we could keep the data in the project files, but not show them in the interface (e.g. by just not showing anything before a certain date or so) - if a topic becomes hot again, we can then just change this - I doubt our users will go off digging really deep down to find old projects

But then we'd have to make sure the actual data is still accurate :) I'd remove it for now. The couple of projects where I saw a contact were either from people I don't know, or ones who left EPFL... (and came back, in the case of Solal :)