Re: tsearch2: plainto_tsquery() with OR?

Thanks for your response! Let me try to elaborate what I meant with my 
original post.

If R is the set of words in the tsvector for a given table row and S is 
the set of keywords to search for (entered by e.g. a website user) I 
would like to receive all rows for which the intersection between R and 
S is nonempty. That is: The row should be return if just there is SOME 
match. S does not necessarily need to be a subset of R.

Furthermore I would like a measure for how "nonempty" the intersection 
is (we would call this measure "the rank").
Example:
For R = "three big houses" and S = "three small houses" the rank should 
be higher than for R = "three big houses" and S = "four small houses" as 
the first case has two words in common while the second case has only one.

A version of plainto_tsquery() with a simple OR operator instead of AND 
would solve this problem somewhat elegant:
1) I can now use the conventional "tsvector @@ tsquery" syntax in my 
WHERE clause as the "@@" operator will return true and thus include the 
row in the result. Example:
   select to_tsvector('simple', 'three small houses')
          @@ 'four|big|houses'::tsquery;
would return "true".

2) The rank() of the @@ operator is automatically higher when there is a 
good match.


An example where this OR-version of plainto_tsquery() could be useful is 
for websites using tags. Each website entry is associated with some tags 
and each user has defined some "tags of interest". The search should 
then return all website entries where there is a match (not necessarily 
complete) with the users tags of interest. Of course the best matching 
entries should be displayed top most.


I find it important that this function is a part of tsearch2 itself as:
1) The user can input arbitrary data. Also potentially harmful data if 
they are not escaped right.
2) Special characters should be stripped in just the same way as 
to_tsvector() does it. E.g. stripping the dot in "Hi . there" but 
keeping it in "web 2.0". Only tsearch2 can do that in a clean consistent 
way - it would be fairly messy if some thirdparty or especially some 
website-developer-homecooked stripping functionality is used for this.