Changes to Search causing BIG problems

Hi! Thanks for taking the time to give us feedback about the new search page.

First, can you help me understand what was broken about the feedback form (report.php)? You mention it has "3 dead links" but it works just fine from my end, so I'm wondering if there may be a login or browser-specific issue?

Let me assure you that the search results are not *random*. Every page is scored for suitability given the page's content, title, and the terms you were searching for. We've just deployed this scoring algorithm and we're expecting to need to do some fair amount of tuning with it to get ideal results - with feedback from customers like you we'll be able to very quickly get you the results that you need. To most speedily and effectively influence the search ranking, specific feedback about what search you did, what you were expecting to see, and what you actually saw are *incredibly* helpful.

(Quick point of clarification - while the results page is not paginated at the present time, the results are not truncated to the first 100 results as suggested.)

Not sure what you mean by inverted commas? If you mean "quotes" then yes, that's the right way to do phrase matching. I think what you're asking for is to have terms that appear next to each other in the text be more highly scored even when quotes are not included, which is fair.

So, specific things for me to work on:
1) Exact matches on pages should always come first (agreed, wise, easy)
2) More highly score terms that appear next to each other, even when not in a quoted phrase search.

Any others?

David,

Many thanks for the detailed response - it's taken me a few days to reply as I wanted to answer each element.

**"Dead links" issue**

Clicking on "Don't like our new Search? Help us make it better."

Leads to (http://suncatest.pbwiki.com/report.ph...) which has the text:

"Home » PBwiki Search Feedback
Report Search Feedback Previously Fixed Issues Known Issues"

Those three terms at the bottom are links but they all link back to the same content-free (except for text above) page:

Report Search Feedback
http://suncatest.pbwiki.com/report.ph...

Previously Fixed Issues
http://suncatest.pbwiki.com/report.ph...

Known Issues
http://suncatest.pbwiki.com/report.ph...

For reference I am browsing my wiki on Mozilla Firefox v. 3.0.1 which is usually ok with pbwiki but does behave differently from IE sometimes (particularly on issues of layout, not usually content).

***Scoring of search result***

Thanks for that information. I assumed this was a new algorithm that was being tinkered with. You mentioned that the title of pages is taken into account and that interests me as my colleague and I have both had a real challenge to search for pages even if using the exact page title.

I would also suggest that a normalised version of the title of pages would be useful to search against as I would perhaps not recall whether there was a hyphen, an underscore or think to replace a space with a %20 etc. if I was searching for a page by title.

***Number of results***

A new message now appears on, it seems, all or most searches. Above the results the text "Note: Due to the large size of your wiki, not all results are included. Please try again later." is shown. This is nice to know but pagination and easier browsing through results would be super. I can think of several ways in which I would prefer priorities to be ranked but roughly I'd think I'd expect results in about this order:

1) pages with (exact) search string in title
2) pages with (exact) search string in tags
3) pages with (exact) search string somewhere on the page
4) pages with all search terms somewhere in the title/tag or page
5) pages with either/or any of the search terms somewhere in the title/tag/page.

***Some transatlantic translation****

yes. Inverted commas = " " = quotes. My American partner and I had a protracted chat about this and we are both sure that our own way is best. But since you are based in the US I shall take her advise and stick to "quotes" next time ;)

***Your list of things to do****

>So, specific things for me to work on:
>1) Exact matches on pages should always come first (agreed, wise, easy)

Sounds super.

>2) More highly score terms that appear next to each other, even when not in a quoted phrase search.

Also sounds ace.

Many thanks again - I'm very happy to give feedback especially if it will help make pbwiki even better! :)

- nicola.

So based on your feedback (and the feedback of others) I've been continuing to improve search in a number of ways, including what we had discussed here.

1) Exact page matches are now put right at the top.

2) The search algorithm will include implicit quotes - i.e. to more highly score search terms that are found next to each other in a document or title, even if you don't enclose the terms in quotes.

Both of these changes should hopefully be live in production by the weekend.

I'm sorry you're seeing the "not all results are included" message. This is an indication that our search algorithm is timing out when searching your wiki. There are some performance improvements pending to help address this -- improving search performance will be an ongoing task for us.

RE: the search feedback page - augh, I just verified this is an issue for *1.0* wikis, which is why I couldn't reproduce it on my 2.0 wikis. You're right that the feedback form is totally broken for 1.0. I'll look into this now. Thanks for pointing this out!

David,

The changes coming this weekend sound super. I'll have a good play with them early next week and post any feedback here.

The \"not all results are included\" message will, I hope, be made fairly irrelevant with the changes to the algorithm (so there may be more matches but hopefully the one we want will rank high on the list and be displayed anyway). At the moment we have just under 1300 pages on our wiki which is quite a lot but not massive I would think but I'm wondering if having an additional filter/option to search only titles (and possibly headings) and tags of pages would be a faster search based on smallish indexes which might be useful for large wikis like ours?

The search feedback issues (yup, on 1.0 wiki - sorry I wasn't clearer on that before) are still there btw.

On a slightly random side note I've noticed a general disparity in where you go when you click for help/to give feedback since the recent embedding of GetSatisfaction and new walkthrough help area.

When you click help on a wiki page it takes you to the radio button list of help options that eventually lead only to a feedback form. However if you click help on the main pbwiki page you end up in this GetSatisfaction area which is often extremely useful even if just for searching through other people's feedback and problems. Is there a reason the walkthrough area (http://pbwiki.com/help.php) doesn't sit at the top of the GetSatisfaction area (http://help.pbwiki.com/helpstart.php) or at least link to it?

Thanks!

David,

Have just been looking over search again on our wiki. Firstly the links to report feedback are still - as per my previous comment- doing nothing.

The search itself seems to be bringing back slightly fewer results (which is good) but the changes to how search terms are prioritized in the results still seems less than fabulous. It seems that they are being prioritised left to right so results including the first word in a searched for phrase is shown first, results featuring multiple words follow, results featuring the last words are shown last. Using inverted commas/quotes does only partly gets around that as this will only search for an exact phrase. The results for specific phrases in inverted commas/quotes does seem to be working substantially better though which is super ;)

Finally it seems that the names of pages are still not being included in searches which is really frustrating. I would expect page names to be indexed as it is obviously not very efficient to locate pages through the All Pages link for large wikis (indeed it is also especially slow and difficult to do this on large wikis) and it is sometimes necessary to search for pages particularly when they are not used very often and/or need a small quick change....

Many thanks,
Nicola.