Feature Proposal: Implement a sorting by relevance in TWiki-Standard-edition

Motivation

The weakest part in TWiki was the search ... until I found out today, that there is an easy solution.

Description and Documentation

Here you can see my initial request with the problem, that the standard search results page is very poor: http://twiki.org/cgi-bin/view/Codev/GoodSortingForSearchResults

Here you can find a solution, that works perfectly!

http://twiki.org/cgi-bin/view/Support/OrderSearchResultsMostRelevantFirst

The only thing, that is needed to be done is, to implement the code in the TWiki-standard-repository.

I believe, that there are no pitfalls or counter-arguments.

This is a true and easy quick win!

Examples

Impact

Implementation

-- Contributors: MartinSeibert - 15 May 2008

Discussion

GO! Please implement it...

One question: What about searching and ranking across multiple webs? If the ranking isn't possible across webs than this is something the average users will complain about for sure...

-- CarloSchulz - 16 May 2008

I welcome the feature. But the relevance ranking should not be displayed as numbers. Google Directory uses a better, more intuitive way. It can easily be implemented with CSS or even with 2 gif images.

-- ArthurClemens - 16 May 2008

I'd be happy to see this used more widely, but then I'm biased

Carlo, this approach works with multiple webs, and puts all results in one table.

A few thoughts:

First, performance could be an issue - it appears that Search.pm goes to some effort to search efficiently. I suspect the context specific searching I added could be faster if it were more tightly integrated into Search. Has anyone tried this on a "big" site?

Second, the context patterns & weightings could be better, and might even be site-specific. (For example, certain webs or templates might be better than others). Keeping this outside of Search.pm probably makes it easier to customize (but hurts performance).

Finally, should column sorting be done client side (in Javascript) or server side. The former is faster and easier on the server, except when there are more results than will fit on one page (another issue for large sites).

-- ClifKussmaul - 16 May 2008

While Javascript based column sorting is useful, we do need static on server sorting too - so that ordered reports can be generated and stored - fo example making a pdf using GenPdfAddon.

Clif - we'd like to see your code especially if it fits into the other Search related work for GeorgeTownRelease - ResultSets etc.

-- SvenDowideit - 17 May 2008

Sven, the changes to Search.pm and a sample AdvancedSearch.txt are in FormattedSearchPatternAndCountContext.

The changes to Search.pm are minor, and localized.

-- ClifKussmaul - 17 May 2008

If performance becomes an issue caching could be of help. But still i believe that this is a tremendous move from a mediocre search result to a good one. And my first experienced in practice are good. The algorithm is rather week for very very long documents. But that is acceptable.

-- MartinSeibert - 17 May 2008