why would search result not order accurately by score?

[johan]

I have a bit of a strange problem on v6 - my search result just won't order by score for a pluralized word. I don't have anything funny going on with zoomboost or weightings or stemming.

I have a particular page dedicated to bellows, a piece of photographic apparatus, always written with an s at the end. Nobody searches for or mentions bellow with any frequency in the photography world. But zoomsoft seems to be having issues with searches for this pluralized word.



On a search for the singular - bellow - a bellows page with a score of 625 is 5th, behind 3 results with lower scores
http://extreme-macro.co.uk/search/sitesearch.php?zoom_query=bellow&sa=Search

On a search for the plural - bellows - a bellows page with a score of 625 is 14th, behind 12 results with lower scores
http://extreme-macro.co.uk/search/sitesearch.php?zoom_query=bellows&zoom_page=2&zoom _per_page=10&zoom_and=1&zoom_sort=0



Anyone shed any light?

Thanks

[johan]

I's almost as if I have 'ensure domain diversity' checked but that's not an option in my 6 (at least there's no setting under search page > )

[johan]

So it turns out that this software is actually pretty useless and has been wrong since the day I started using it. It seems that the ordering is by the number of mentions of a word on the page, but that this software simply misses out and doesn't seem to be able to see those mentions.

[David]

You are posting in the V6 forum. We no longer support V6 of the software. So you might want to have a look at V7 instead.

I don't have anything funny going on with zoomboost or weightings or stemming.

You do seem to have stemming turned on. This why a search for bellows and bellow return the same set of results.
You can turn this option off if you want. But we don't normally suggest it unless it is causing a serious problem. The behaviour seems to be OK here.

You have turned off the display of the score. So we can't verify any of the above.

For both searches the top 10 results seem highly relevant to the search time. So I am not sure what the problem is? Or why you think it is wrong?

The score (and subsequent ranking) of pages is described here
https://www.wrensoft.com/zoom/support/faq_score.html
It isn't just based on the number of word occurrences.

[johan]

Thank you for the reply! I've turned the scoring etc back on - my strange result is at http://extreme-macro.co.uk/search/sitesearch.php?zoom_query=bellows&sa=Search

To me it looks like it is ordered by terms matched first, but there's something strange going on.

Result 15 on page 2 has a score of 554, surely this should be the third result on page one, below tilt shift bellows which has the score of 595?

Thank you for your help

[Ray]

Hi Johan,

As you noticed, the "Terms matched" has greater preference over the "Score".

So result 15 in your last example has "Terms matched: 2, Score: 554" and is thus on Page 2 of results. Every result on Page 1 has "Terms matched" 3 or more.

So the scoring is behaving as expected, albeit there is some misunderstanding.

Now the next question is why does a single search word of "bellow" lead to multiple terms matching.

Looking at your results, I think you have enabled "Substring match for all searches" (under "Configure"->"Languages"). That's why I also see matches for "bellows-macro-experiment", "custombellows", etc.

This is NOT a recommended setting for English. It is only designed for languages which are not space delimited (like Japanese).

Many people think they want this behaviour at a quick glance because it seems like an easy solution to getting some keywords to match in particular scenarios. But it really doesn't work in most cases and hamper the usefulness of your search engine.

For example, any one who searches for "Mac" on your site are getting "Macro". "Cat" would match "category" or "concatenation", etc. It makes searching for many words impossible.

Instead, there are options like the Word Rules ("Configure"->"Indexing options"->"Indexing word rules") which allow you to specify what characters make or break a word.

If you uncheck "hyphens" from joining words, then an example like:
bellows-macro-experiment

Is broken up into 3 words:
bellows
macro
experiment

And you can search for any of those words and it will match accordingly.

The stemming feature ("Configure"->"Languages") allows "bellows" to match "bellow" and vice versa.

So I would advise turning off the "Substring match" feature and configuring the above.

Hope that helps.

[johan]

Ray, thank you so much for taking the time to write all this up for me.

You are correct I did indeed have "Substring match for all searches" enabled and I'm doing another run now without it checked.

Again you're also correct about the misunderstanding regarding the scoring because the page that measures "Terms matched: 2, Score: 554" actually mentions the terms a lot more than 2 times.

It maybe that respidering it with "Substring match for all searches" corrects this, I'll do this and re-evaluate.

Thank you again!!!

[johan]

Fabulous, this has sorted it.

Clearly my understanding of "terms matched" was wrong as well, as now that there's no term matching going on, all results equate to one term matched and being so, the results are no ordered by score within terms matched. Which makes a whole load of sense and is the correct behaviour I'd expect ... so thank you again

[kaufenpreis]

This software simply misses out and doesn't seem to be able to see those mentions.

[Ray]

kaufenpreis -- can you elaborate what you mean and which scenario you are talking about.