Search does not always work

[D_Hailsham]

I have come across two problems with the search facility.

1. The advanced search for an exact phrase does not work properly.

For example: a search for "set abv to pump head", which occurs in
//www.diynot.com/forums/viewtopic.php?p=729215#729215, does not find this post. Instead it only finds two posts containing the individual words, i.e. it does an AND query.

Incidentally my purpose was to find the original post from which this phrase originated.

2. searches for posts by particular members does not work.

For example: a search for honeywell in the plumbing forum which I have posted, i.e author d_hailsham. returns:

Your search - ((honeywell) forum_id:7) author:"d_hailsham" - did not match any documents.

But if I do a Forum search by author for d_hailsham, it returns all my posts.

 

[DIYnot]

Hi,

The reason your posts are not appearing is because they haven't been indexed yet.

Normally re-indexing of posts happens on a weekly basis, but where we have been adjusting things and moving things around the re-indexing process has been suspended. It has actually started again today and the re-indexing of the main forum is under way as I type. This will probably finish some time during the night. The re-indexing should return to normal and within the week be back up to date and hopefully re-running your queries will return the results.

Apologies for the inconvenience in the mean time.

 

[D_Hailsham]

Thanks for the info about re-indexing, but it does not really explain why an Exact Phrase query is handled as a string of AND queries. I have also observed the problem with searches for posts by a particular author over a period of time, so it is nothing new.

Perhaps you could confirm something else relating to searches. Would it be correct to assume that you have a stop list of common words, e.g: the; and; not etc which are not indexed and therefore cannot be searched? Does this also apply if they occur in an exact phrase searches?

 

[DIYnot]

I'm not too sure about what you mean, when I enter your query into the Exact phrase query it searches as an exact phrase - i.e. inside "exact phrase", what process are you going through to observe this behaviour?

Regarding stop list, there is one, and they are filtered from both posts when indexed and from queries. So, an exact phrase query shouldn't fail if it contains a stop word because it should be filtered out, as you can observe in this example, the only side effect is that the stop words aren't highlighted.

 

[D_Hailsham]

I have a feeling that you idea of an exact phrase search may be different from mine. In my book, a phrase search, searches only for the exact string of the complete phrase. It does not search for the individual words of the phrase, which is nothing more than an AND search.

In your example, the first result does contain the phrase "out the electrics to the house" but it also highlights contains the word electrics where it is not contained in the phrase requested.

In the second result, the phrase searched for does not actually occur, instead it finds: out the electrics in the house

Your search is therefore doing no more than an AND search.

The problem, of course, is that a proper phrase search is really a Full Text search; and indexes and stop lists are irrelevant in this case.

In my search, I was looking for "set abv to pump head"

The search returned:

Your search - "set abv to pump head" - did not match any documents

Another search for "determine head loss", which occurs in the same post, returns:

Your search - "determine head loss" - did not match any documents.

Two other things that I have noted are:

1. there is no way of searching the titles of the post
2. results are presented in a haphazard order. They are not in topic, date poster or any other order that I can see.

Lastly, and I have raised this before, could we have just one result returned from each topic (say the one which contains the most words sought - many search engines give a percentage hit), rather than the many results per topic which occurs at present?

Is there any way of providing these facilities?

 

[DIYnot]

Sorry, I see what you're getting at now

A topic based search is planned and will be available in due course, the same with searching by title, I'm not too sure on time scales at this point in time.

The ordering is based on a score which is derived from the frequency of the query words appearing in a post.

 

[D_Hailsham]

The ordering is based on a score which is derived from the frequency of the query words appearing in a post.

That explains everything