Wildcard search and spelling suggestions

[quintic]

I have "Provide spelling suggestions when less than 2 results are found" set. So I was surprised when a search for flexib* returned five results and a list of spelling suggestions! Worse, one of the suggested words was "flexible", which actually accounted for four of the five matches.

Checking the PHP, it appears the check for less than two matches is performed for each word found that matches the wildcard. I don't think that makes a lot of sense.

I'm going to modify the PHP myself to issue wildcard suggestions only when the total number of results is less than $SpellingWhenLessThan, but I'd appreciate it if this could be added to a list so that the CGI version can be fixed. (Assuming you agree it's an issue! If not, then the indexer option should be modified to make clear that this is per word behaviour.)

Finally, is the C++ source for the CGI version available?

Nick

[Ray]

Yes, the spelling suggestion is determined per search term. This is so that we can offer suggestions even when only one of the words are incorrectly spelt (a search for "cat doggs" would be offered "cat dog").

We may change the description in the Configuration window to something like "Provide spelling suggestions when less than x results are found for a search term".

We are not currently offering the C++ source code for the CGI version.

However, we have plans to introduce a SDK license which would include the CGI source code, amongst additional documentation and specifications for other components of Zoom (such as the ZCFG file, etc.). Contact us if you are interested in this.

[quintic]

Yes, the spelling suggestion is determined per search term. This is so that we can offer suggestions even when only one of the words are incorrectly spelt (a search for "cat doggs" would be offered "cat dog").

I can see the value of that, especially when 'Default to "match all search words"' is not checked. However, I don't think it makes sense when the search term is a wildcard.

Surely a search for flexib* should not trigger spelling suggestions merely because one of the matching words happens to match fewer than $SpellingWhenLessThan times? (Yet total results >= $SpellingWhenLessThan.) Unlike the "cat doggs" case above, the user has no control over each matching word, only over the total number of matches. (Consider a search for flexxib* vs. flexib*) Therefore it should be the total number of matches that triggers spelling suggestions.

This raises another question: should a spelling suggestion for a wildcard always be another wildcard? Very tricky!

Nick

[Ray]

Surely a search for flexib* should not trigger spelling suggestions merely because one of the matching words happens to match fewer than $SpellingWhenLessThan times? (Yet total results >= $SpellingWhenLessThan.) Unlike the "cat doggs" case above, the user has no control over each matching word, only over the total number of matches. (Consider a search for flexxib* vs. flexib*) Therefore it should be the total number of matches that triggers spelling suggestions.

I did a few tests and I see the problem you're referring to, although it's not quite the same as what you described and the solution is somewhat different to your suggestion.

The problem is not due to Zoom looking at each search term to determine whether suggestions are needed - this is fine. The problem is that there's a bug in the script which causes Zoom to miscount the number of results for a wildcard search term. In fact, it only counts the last wildcard term matched, which means that sometimes, it thinks there were far less search results returned for a single wildcard search term, than there really was. We'll fix this in the next release, and it should subsequently address your initial problem with the suggestions feature behaving not how you expected it to.

[quintic]

Thanks. I see what you mean, now that I've run the program with a few debug print statements. For now, I've set $SpellingWhenLessThan to 1, which effectively switches off spelling suggestions, for both regular and wildcard searches, unless no matches are found.