search results should be sorted by relevancy

[OriHoch]

We made a lot of changes to search functionality, but neglected to check if results are returned by correct relevancy scoring

AKA - most relevant results should be displayed first

see bug #185 for an example of 1 such problem

TODO

1. research and review the existing relevancy rules - what works / what doesn't work / how it should work
2. implement the results of this research, either all of them, or phase them in stages

[OriHoch]

see the comments on #185 for some implementation notes
I think the most simple solution is to define ordering of collections - in that case, just need to decide what's the order of the collections (e.g. places first, then what order?)

[TheGrandVizier]

I think we should go with the trends we are seeing from analytics. As in, Countries get the most exposure, etc.
@Nurit any input on this?
This will often be the first result screen they see.

[TheGrandVizier]

@Netush

[OriHoch]

@TheGrandVizier I think we have a long way to go until we will need to use analytics data for ranking

let's do simple things first, like sorting the collections - which will solve the Germany problem (#185)

In any case, try to describe examples of specific problems which will be solved by your suggestions - (like the Germany problem)

[OriHoch]

opened an issue to handle the simple fixes - #187

please continue working on this issue for long-term / better solutions / research etc.

[Netush]

It’s a problem to decide on the ordering of the collection because we want the most relevant item to appear first: when you search “Germany” it should be places, but when you search “Cohen” it should be family name.
Therefore I suggest not to determine the order of the collection but that results from family trees never be in the first 2 rows (6 results) - only if we don’t have other items of course

[OriHoch]

I think I might have understood the source of confusion.. when I say ordering of collections I don't mean that places will always appear before persons

I mean that the search engine internally will have an ordering of collections (based on our knowledge of the data) and will use that ordering to determine the relevancy

more precisely, we can assign a "boost value" to each collection - which the search engine will use to determine relevancy

it could be as simple as saying the persons are at the bottom of the order and order of other collections is irelevant. Or more complex, like saying the places have boost of 10, persons will have a boost of 1 and other collections with different boosts in between

how the boosts will affect the final results is hard to tell, we will have to test and tweak it