Johann Spies <johann.spies@gmail.com> writes:
> I have a record with this in the 'ti' field:
> On type A and type B uncertainties and its propagation without
> derivatives: a contribution to incorporate contemporary metrology to
> Physics' laboratories in higher education
> and tsv contains the following for this record:
> "'activ':129 'altern':73,94 'assumpt':62 'b':6,39 'concept':28,122
> 'contemporari':17 'contribut':14 'deriv':12,91 'discuss':42
> 'earliest':127 'educ':24,133 'estim':76 'evalu':34 'experiment':128
What have you got default_text_search_config set to? Not "simple":
regression=# select to_tsvector('simple','higher education');
to_tsvector
--------------------------
'education':2 'higher':1
(1 row)
regression=# select to_tsvector('english','higher education');
to_tsvector
---------------------
'educ':2 'higher':1
(1 row)
> When I use the query
> select ut, ti
> from isi.ritem A
> where
> A.tsv @@ to_tsquery('Simple','higher & education')
> The result is zero.
> But then I use
> A.tsv @@ to_tsquery('Simple','higher & educ')
> I get more than 54000 records as result.
Not surprising: the simple config doesn't assume the lexemes are English
words.
regression=# select to_tsquery('Simple','higher & education');
to_tsquery
------------------------
'higher' & 'education'
(1 row)
regression=# select to_tsquery('english','higher & education');
to_tsquery
-------------------
'higher' & 'educ'
(1 row)
> That makes me think the full text search is probably not very reliable.
If misused, yeah. You need to have a fairly clear idea of which language
you are indexing; or maybe I should just say that you need to use a
consistent text search configuration selection. (Words in other languages
will still work, they just might not get stemmed usefully.)
regards, tom lane
