Need to optimize query performance

1 view
Skip to first unread message

Mayur Barge

unread,
Aug 1, 2016, 5:05:49 AM8/1/16
to sta...@clarkparsia.com
Hello,

I need to optimize performance of below query to bring down the execution time.
Query plan is attached below.

Query
select DISTINCT ?orgLabel FROM NAMED <http://namedgraph/autosuggest/> where
{
GRAPH ?g {
?urn smt:capitalCaseLabel ?label ;
     smt:orgLabel ?orgLabel .

  ?s smt:searchLabel ?urn .

FILTER(STRSTARTS(?label,"boi"))

}
}
order by (STRLEN(?orgLabel)) ?orgLabel
limit 10

The Query Plan:

From Named <http://namedgraph/autosuggest/>
Slice(offset=0, limit=10) [cardinality=10]
  OrderBy(ASC(STRLEN(?orgLabel)), ASC(?orgLabel)) [cardinality=862K]
    Distinct [cardinality=862K]
      Projection(?orgLabel) [cardinality=862K]
        MergeJoin[?urn] [cardinality=862K]
          Filter(STRSTARTS(?label, "boi")) [cardinality=431K]
            Scan[PSO](?urn, <http://namedgraph/terms/capitalCaseLabel>, ?label){?g} [cardinality=862K]
          MergeJoin[?urn] [cardinality=862K]
            Scan[PSO](?urn, <http://namedgraph/terms/orgLabel>, ?orgLabel){?g} [cardinality=862K]
            Scan[POS](?s, <http://namedgraph/terms/searchLabel>, ?urn){?g} [cardinality=862K]

Thank you,
Mayur

Pavel Klinov

unread,
Aug 1, 2016, 5:32:58 AM8/1/16
to sta...@clarkparsia.com
It seems that your question is essentially a duplicate of [1]. Please try the suggestion given there.

Best,
Pavel


On Mon, Aug 1, 2016 at 11:05 AM, Mayur Barge <mayurb...@gmail.com> wrote:
Hello,

I need to optimize performance of below query to bring down the execution time.
Query plan is attached below.

Query
select DISTINCT ?orgLabel FROM NAMED <http://materials.springer.com/autosuggest/> where
{
GRAPH ?g {
?urn smt:capitalCaseLabel ?label ;
     smt:orgLabel ?orgLabel .

  ?s smt:searchLabel ?urn .

FILTER(STRSTARTS(?label,"boi"))

}
}
order by (STRLEN(?orgLabel)) ?orgLabel
limit 10

The Query Plan:

Slice(offset=0, limit=10) [cardinality=10]
  OrderBy(ASC(STRLEN(?orgLabel)), ASC(?orgLabel)) [cardinality=862K]
    Distinct [cardinality=862K]
      Projection(?orgLabel) [cardinality=862K]
        MergeJoin[?urn] [cardinality=862K]
          Filter(STRSTARTS(?label, "boi")) [cardinality=431K]
            Scan[PSO](?urn, <http://materials.springer.com/terms/capitalCaseLabel>, ?label){?g} [cardinality=862K]
          MergeJoin[?urn] [cardinality=862K]
            Scan[PSO](?urn, <http://materials.springer.com/terms/orgLabel>, ?orgLabel){?g} [cardinality=862K]
            Scan[POS](?s, <http://materials.springer.com/terms/searchLabel>, ?urn){?g} [cardinality=862K]

Thank you,
Mayur

--
-- --
You received this message because you are subscribed to the C&P "Stardog" group.
To post to this group, send email to sta...@clarkparsia.com
To unsubscribe from this group, send email to
stardog+u...@clarkparsia.com
For more options, visit this group at
http://groups.google.com/a/clarkparsia.com/group/stardog?hl=en

Mayur Barge

unread,
Aug 1, 2016, 5:41:46 AM8/1/16
to Stardog
Hello Pavel,

The link you shared was a question asked by one of my colleague. We do not have the option of having full text search.
We need to show results as user types in words, hence we require that partial matches should be displayed.

Thank you,
Mayur

Pavel Klinov

unread,
Aug 1, 2016, 5:48:33 AM8/1/16
to sta...@clarkparsia.com
What is the exact reason that you cannot use a query with a full-text search predicate but can use a query with FILTER(STRSTARTS)? You can use wildcards as well as other Lucene search modifiers in full-text queries.

Best,
Pavel

Mayur Barge

unread,
Aug 1, 2016, 6:54:03 AM8/1/16
to Stardog
With full-text search we are getting the results wherein word is matched in between. As shown in Blue
We have a requirement wherein terms starting with ben should be displayed.

For example 
benzene
abc benzene
But not the words wherein ben comes in between, words such as
chlorobenzene


select ?l FROM <http://namedgraph/autosuggest/> where
{
 ?s ?p ?l.
  
 ?l <tag:stardog:api:property:textMatch> ('ben' 10).

}
order by (STRLEN(?label)) ?label
limit 10

Results

BeN<sub>2</sub>O<sub>6</sub>
BeN<sub>2</sub>O<sub>6</sub>
BeN<sub>2</sub>O<sub>6</sub>
BeN<sub>2</sub>O<sub>6</sub>
Ben-Hex
Ben-Hex
C<sub>32</sub>H<sub>16</sub>BeN<sub>8</sub>
C<sub>32</sub>H<sub>16</sub>BeN<sub>8</sub>
C<sub>16</sub>H<sub>32</sub>BeN<sub>2</sub>
C<sub>16</sub>H<sub>32</sub>BeN<sub>2</sub>

Thanks,
Mayur

Zachary Whitley

unread,
Aug 1, 2016, 7:29:29 AM8/1/16
to sta...@clarkparsia.com
Try making it a wild card query. "Ben*". You might want to look into how Lucene analyzers work if you aren't already familiar with them and specifically how Stardog uses them. http://docs.stardog.com/#_user_defined_lucene_analyzer
You received this message because you are subscribed to the Google Groups "Stardog" group.
To unsubscribe from this group and stop receiving emails from it, send an email to stardog+u...@clarkparsia.com.

Mayur Barge

unread,
Aug 1, 2016, 8:13:41 AM8/1/16
to Stardog

When we tried query having filter we got below results

select DISTINCT ?orgLabel FROM <http://namedgraph/autosuggest/> where
{
   ?urn smt:modLabel ?label ;
      smt:orgLabel ?orgLabel .

 ?s smt:searchLabel ?urn .

 FILTER(STRSTARTS(?label,"ben"))

}
order by (STRLEN(?label)) ?label
limit 10

Results
Bena
BeNC
BeNi
Benzac
Benzen
benzil


We got four results having word length four.

But when we tried with wild card query on full-text search, we got single result with word length four.

select DISTINCT ?orgLabel FROM <http://namedgraph/autosuggest/> where
{

  ?label <tag:stardog:api:property:textMatch> ('ben*' 10000).

   ?urn smt:modLabel ?label ;
      smt:orgLabel ?orgLabel .

 ?s smt:searchLabel ?urn .
}
order by (STRLEN(?label)) ?label
limit 10

Results
BeNi
benzil
Benzil
BeNi<sub>2</sub>Si

We need to search all records and then filter out those ones which matches "ben" and order by minimum word length.

Thanks,
Mayur

Marcel Karnstedt-Hulpus

unread,
Aug 2, 2016, 12:53:17 AM8/2/16
to sta...@clarkparsia.com
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA512

If we increase the limit for the Lucene query even more, we find the
shorter results like Bena and BeNC as well. So it seems the core issue
is that Lucene's ranking doesn't fit our purpose. In fact, all
returned labels have exactly the same score.

It would be great if we could adapt the scoring to our needs taking
string length into account, which might also render the application of
STRLEN() unnecessary.

Is there a way to do this with Stardog's Lucene similar to how we
could adapt the used Analyzer?


Thanks,
Marcel
> <javascript:>> wrote:
>
>> With full-text search we are getting the results wherein word is
>> matched in between. As shown in Blue We have a requirement
>> wherein terms starting with ben should be displayed.
>>
>> For example benzene abc benzene But not the words wherein ben
>> comes in between, words such as chlorobenzene
>>
>>
>> select ?l FROM <http://namedgraph/autosuggest/
>> <http://namedgraph/autosuggest/>> where { ?s ?p ?l.
>>
>> ?l <tag:stardog:api:property:textMatch> ('ben' 10).
>>
>> } order by (STRLEN(?label)) ?label limit 10
>>
>> Results
>>
>> BeN<sub>2</sub>O<sub>6</sub> BeN<sub>2</sub>O<sub>6</sub>
>> BeN<sub>2</sub>O<sub>6</sub> BeN<sub>2</sub>O<sub>6</sub>
>> Ben-Hex Ben-Hex C<sub>32</sub>H<sub>16</sub>*BeN*<sub>8</sub>
>> C<sub>32</sub>H<sub>16</sub>*BeN*<sub>8</sub>
>> C<sub>16</sub>H<sub>32</sub>*BeN*<sub>2</sub>
>> C<sub>16</sub>H<sub>32</sub>*BeN*<sub>2</sub>
>> sta...@clarkparsia.com <javascript:> To unsubscribe from this
>> group, send email to stardog+u...@clarkparsia.com <javascript:>
>> For more options, visit this group at
>> http://groups.google.com/a/clarkparsia.com/group/stardog?hl=en
>> <http://groups.google.com/a/clarkparsia.com/group/stardog?hl=en>
>> --- You received this message because you are subscribed to the
>> Google Groups "Stardog" group. To unsubscribe from this group and
>> stop receiving emails from it, send an email to
>> stardog+u...@clarkparsia.com <javascript:>.
>
> -- -- -- You received this message because you are subscribed to
> the C&P "Stardog" group. To post to this group, send email to
> sta...@clarkparsia.com To unsubscribe from this group, send email
> to stardog+u...@clarkparsia.com For more options, visit this
> group at
> http://groups.google.com/a/clarkparsia.com/group/stardog?hl=en ---
> You received this message because you are subscribed to the Google
> Groups "Stardog" group. To unsubscribe from this group and stop
> receiving emails from it, send an email to
> stardog+u...@clarkparsia.com
> <mailto:stardog+u...@clarkparsia.com>.

- --
Dr. Marcel Karnstedt-Hulpus
www.marcel.karnstedt.com

Senior Data Architect
IT & Production Corporate Markets / Databases
Springer Nature
www.springernature.com

tel +49-6221-4878868
mob +49-151-23125052
Tiergartenstrasse 17, 69121 Heidelberg, Germany

- ---
Springer Nature is a major new force in scientific, scholarly,
professional and educational publishing, created in May 2015 through
the combination of Nature Publishing Group, Palgrave Macmillan,
Macmillan Education and Springer Science+Business Media.
- ---
Branch of Springer-Verlag GmbH, Heidelberger Platz 3,
14197 Berlin, Germany
Registered Office: Berlin / Amtsgericht Berlin-Charlottenburg,
HRB 91881 B
Directors: Derk Haank, Martin Mos, Petrus W. J. Hendriks
- ---
- ---
-----BEGIN PGP SIGNATURE-----
Comment: GPGTools - https://gpgtools.org

iQEcBAEBCgAGBQJXoCc4AAoJELjmZAjIQprmj6sIALNzbjxyns2eAmYyLVt1liIg
jDDPQzoJ1yiZkS+a0cpAju1LY1CGVLOBzR0JNSnGW9w33PVSpGyZ5kSWPneEDkJA
2UyYyQ5a804b8q8+nrHPSzd+RhAPDTYIIYg8AjOFOjMqFYekPMeqtgXkhdjhyzvT
fLbeZpHMqDkeDTu1bHAQPdYYLuAYAudOGK16jgKZD/JLmz9TPz1KIm7rbaob3maJ
tqOH3nV+SdLmAE8NQptEJTtJmhjcFk1VU9hTq0aj4nDcs4POyi3vgjT5PtmdoxfB
MKAGhwC1gbXkOr8yQf8kVe1s3W3kormENl2Td6stE269VXmmCj5hL0YM2mKlAI8=
=wY0/
-----END PGP SIGNATURE-----

Pavel Klinov

unread,
Aug 2, 2016, 5:48:38 AM8/2/16
to sta...@clarkparsia.com
Hi Marcel,

I think using wildcards turns Lucene queries into constant-score queries so by using wildcards you essentially throw away the ranking. 
I'll need to double check though (I'm no Lucene expert).

The main performance bottleneck in such queries is that FILTERs with predicates such as STRSTARTS are applied to a lot of literals.
If you can use a full-text predicate in the same query which would cut down the number of literals, e.g. to those having 'Ben' somewhere,
then applying the FILTER on top of that should be faster (as well as ordering the results). 
The Stardog query optimizer should first apply the search predicate and then FILTERs in such cases. 
So the trick is to use a full-text predicate which will match a superset of what you need.

Can't say much about custom scoring at this point but we can look at that in the future.

Best,
Pavel

Marcel Karnstedt-Hulpus

unread,
Aug 2, 2016, 7:10:54 AM8/2/16
to sta...@clarkparsia.com
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA512

Thanks for the suggestions Pavel. Unfortunately, it doesn't seem to
work for us.

Filtering first for

<tag:stardog:api:property:textMatch> ('ben')

returns only results having 'ben' as a separate term, not those
containing it as a sub-string:

Ben-Hex
Ben-u-ron
BeN &lt;g&gt;

I'm also far from being a Lucene expert, might need to dig a bit
deeper for alternative syntax. Any suggestions from your team are very
welcome.


Two side notes:

1. The initial query plan is also not as expected:

Explaining Query:

select DISTINCT ?orgLabel ?score FROM <http://graph/autosuggest/> where
{
(?label ?score) <tag:stardog:api:property:textMatch> ('ben' 100000).
FILTER(STRSTARTS(?label,"ben")) .
?urn smt:modLabel ?label ;
smt:orgLabel ?orgLabel .
}
order by (STRLEN(?label)) ?label
limit 10

The Query Plan:

- From <http://materials.springer.com/autosuggest/>
Slice(offset=0, limit=10) [cardinality=10]
Distinct [cardinality=50K]
Projection(?orgLabel, ?score) [cardinality=50K]
OrderBy(ASC(STRLEN(?label)), ASC(?label)) [cardinality=50K]
MergeJoin[?urn] [cardinality=50K]
Sort(?urn) [cardinality=50K]
MergeJoin[?label] [cardinality=50K]
Sort(?label) [cardinality=50K]
Filter(STRSTARTS(?label, "ben")) [cardinality=50K]
Full-Text(query='ben', limit=100000) ->
(results=?label, scores=?score) [cardinality=100K]
Filter(STRSTARTS(?label, "ben")) [cardinality=431K]
Scan[POS](?urn,
<http://materials.springer.com/terms/modLabel>, ?label) [cardinality=862K]
Scan[PSO](?urn,
<http://materials.springer.com/terms/orgLabel>, ?orgLabel)
[cardinality=862K]

The STRSTARTS filter is apparently applied twice and the results
joined afterwards. Stardog version 4.1.1. Encapsulating both filters
in a sub-query solves this. Still not the wanted results, though.


2. I think I might have mentioned that before: full-text search
returns 0 results once I put it *alone* into a sub-query. Some issue
about binding variables from full-text search?


Cheers,
Marcel



On 02/08/16 15:18, Pavel Klinov wrote:
> Hi Marcel,
>
> I think using wildcards turns Lucene queries into constant-score
> queries so by using wildcards you essentially throw away the
> ranking. I'll need to double check though (I'm no Lucene expert).
>
> The main performance bottleneck in such queries is that FILTERs
> with predicates such as STRSTARTS are applied to a lot of
> literals. If you can use a full-text predicate in the same query
> which would cut down the number of literals, e.g. to those having
> 'Ben' somewhere, then applying the FILTER on top of that should be
> faster (as well as ordering the results). The Stardog query
> optimizer should first apply the search predicate and then FILTERs
> in such cases. So the trick is to use a full-text predicate which
> will match a superset of what you need.
>
> Can't say much about custom scoring at this point but we can look
> at that in the future.
>
> Best, Pavel
>
> On Tue, Aug 2, 2016 at 6:53 AM, Marcel Karnstedt-Hulpus
> <marcel.k...@springernature.com
> <mailto:mayurb...@gmail.com>
>>> to sta...@clarkparsia.com <mailto:sta...@clarkparsia.com> To
> unsubscribe from this group, send email
>>> to stardog+u...@clarkparsia.com
> <mailto:stardog%2Bu...@clarkparsia.com> For more options, visit
> this
>>> group at
>>> http://groups.google.com/a/clarkparsia.com/group/stardog?hl=en
>>> <http://groups.google.com/a/clarkparsia.com/group/stardog?hl=en>
>>>
>>>
>>>
>>>
- -- -- -- You received this message because you are subscribed to
>>> the C&P "Stardog" group. To post to this group, send email to
>>> sta...@clarkparsia.com <mailto:sta...@clarkparsia.com> To
> unsubscribe from this group, send email
>>> to stardog+u...@clarkparsia.com
> <mailto:stardog%2Bu...@clarkparsia.com> For more options, visit
> this
>>> group at
>>> http://groups.google.com/a/clarkparsia.com/group/stardog?hl=en
>>> <http://groups.google.com/a/clarkparsia.com/group/stardog?hl=en>
>>>
>>>
>>>
>>>
- -- -- -- You received this message because you are subscribed to
>>> the C&P "Stardog" group. To post to this group, send email to
>>> sta...@clarkparsia.com <mailto:sta...@clarkparsia.com>
> <javascript:> To unsubscribe from this
>>> group, send email to stardog+u...@clarkparsia.com
> <mailto:stardog%2Bu...@clarkparsia.com> <javascript:>
>>> For more options, visit this group at
>>> http://groups.google.com/a/clarkparsia.com/group/stardog?hl=en
>>> <http://groups.google.com/a/clarkparsia.com/group/stardog?hl=en>
>>>
>>>
- --- You received this message because you are subscribed to the
>>> Google Groups "Stardog" group. To unsubscribe from this group
>>> and stop receiving emails from it, send an email to
>>> stardog+u...@clarkparsia.com
> <mailto:stardog%2Bu...@clarkparsia.com> <javascript:>.
>
>> -- -- -- You received this message because you are subscribed to
>> the C&P "Stardog" group. To post to this group, send email to
>> sta...@clarkparsia.com <mailto:sta...@clarkparsia.com> To
> unsubscribe from this group, send email
>> to stardog+u...@clarkparsia.com
> <mailto:stardog%2Bunsu...@clarkparsia.com> For more options,
> visit this
>> group at
>> http://groups.google.com/a/clarkparsia.com/group/stardog?hl=en
>> --- You received this message because you are subscribed to the
>> Google Groups "Stardog" group. To unsubscribe from this group and
>> stop receiving emails from it, send an email to
>> stardog+u...@clarkparsia.com
> <mailto:stardog%2Bunsu...@clarkparsia.com>
>> <mailto:stardog+u...@clarkparsia.com
> <mailto:stardog%2Bunsu...@clarkparsia.com>>.
>
>
> -- -- -- You received this message because you are subscribed to
> the C&P "Stardog" group. To post to this group, send email to
> sta...@clarkparsia.com <mailto:sta...@clarkparsia.com> To
> unsubscribe from this group, send email to
> stardog+u...@clarkparsia.com
> <mailto:stardog%2Bunsu...@clarkparsia.com> For more options,
> visit this group at
> http://groups.google.com/a/clarkparsia.com/group/stardog?hl=en
>
>
iQEcBAEBCgAGBQJXoH+6AAoJELjmZAjIQprmg8IIAKHNaHr2tlAdURtR1zBUVsNu
6JJKJ31PXsRP6huFscRRalPRfet0xLgHoT4n//yhbx0A5JEIABiMNP+zxJ07LyPH
uSDSQzGySxZWtQLdNZmHJqGEOMJfkqOxTtfzjpg7xpeRTxcUNXbuKtdmKL8ccbbm
j/zrkvU5STv3+wi7b3WarSEL+Pq+bQgfjSIuINm7YO1O5G+k7PynpU0qAkQeYXPD
itXhGwUJC+JSTKVvBbW+8aI3lewbWsGmiSN6EJPXbUlyrNIXmLKADNFGWG18h6zQ
F19ceZdSW2FoysUISDrmxOAoCUVeBzTKmNpFLeqN6xJ61RSmIebYOkxxdxmYRvs=
=N4a3
-----END PGP SIGNATURE-----

Pavel Klinov

unread,
Aug 2, 2016, 7:22:45 AM8/2/16
to sta...@clarkparsia.com
On Tue, Aug 2, 2016 at 1:10 PM, Marcel Karnstedt-Hulpus <marcel.k...@springernature.com> wrote:
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA512

Thanks for the suggestions Pavel. Unfortunately, it doesn't seem to
work for us.

Filtering first for

<tag:stardog:api:property:textMatch> ('ben')

returns only results having 'ben' as a separate term, not those
containing it as a sub-string:

Ben-Hex
Ben-u-ron
BeN &lt;g&gt;

I'm also far from being a Lucene expert, might need to dig a bit
deeper for alternative syntax. Any suggestions from your team are very
welcome.

I thought from the previous email that this is where wildcards help (if you don't use scoring), no?
Yes, sometimes Stardog aggressively pushes filters to both operands of a join if it thinks it'd make the join cheaper.
However this logic was changed a bit for 4.1.2 and I don't think it'd do it in this specific case. 
 

2. I think I might have mentioned that before: full-text search
returns 0 results once I put it *alone* into a sub-query. Some issue
about binding variables from full-text search?

Hm, I can't recall. Can you show the exact query so I can try to reproduce?

Thanks,
Pavel

Marcel Karnstedt-Hulpus

unread,
Aug 2, 2016, 1:47:23 PM8/2/16
to sta...@clarkparsia.com
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA512

Hi Pavel,

in-lines...

On 02/08/16 16:52, Pavel Klinov wrote:
>
>
> On Tue, Aug 2, 2016 at 1:10 PM, Marcel Karnstedt-Hulpus
> <marcel.k...@springernature.com
> <mailto:marcel.k...@springernature.com>> wrote:
>
> Thanks for the suggestions Pavel. Unfortunately, it doesn't seem
> to work for us.
>
> Filtering first for
>
> <tag:stardog:api:property:textMatch> ('ben')
>
> returns only results having 'ben' as a separate term, not those
> containing it as a sub-string:
>
> Ben-Hex Ben-u-ron BeN &lt;g&gt;
>
> I'm also far from being a Lucene expert, might need to dig a bit
> deeper for alternative syntax. Any suggestions from your team are
> very welcome.
>
>
>> I thought from the previous email that this is where wildcards
>> help (if you don't use scoring), no?
Yeah, agreed - but if I use 'ben*' then I don't need any additional
filter any more, because it will already return only terms *starting*
with 'ben'. But I still need to fetch quite a large number of
full-text hits, since otherwise I don't get the short candidates. I
presume they are always returned in Lucene-score order, no matter if I
intend to use the score or not.

Sorry, that might be a bit cryptic. Examples:

select DISTINCT ?orgLabel ?score FROM <http://graph/autosuggest/> where
{
{
select ?urn ?label ?score where {
(?label ?score) <tag:stardog:api:property:textMatch> ('ben*' 100000).
FILTER(STRSTARTS(?label,"ben")) .
?urn smt:modLabel ?label .
}
}
?urn smt:orgLabel ?orgLabel .
}
order by (STRLEN(?label)) ?label
limit 10

- -> STRSTARTS filter is unnecessary, performance not satisfying (as
expected)


select DISTINCT ?orgLabel ?score FROM <http://graph/autosuggest/> where
{
(?label ?score) <tag:stardog:api:property:textMatch> ('ben*' 100000).
?urn smt:modLabel ?label .
?urn smt:orgLabel ?orgLabel .
}
order by (STRLEN(?label)) ?label
limit 10

- -> performance not satisfying, since I have to fetch too many text
matches in order to get all results I expect (i.e., doesn't work with
10000 rather than 100000)
ok, will turn on the search feature on our 4.1.2 and test

>
>
>
> 2. I think I might have mentioned that before: full-text search
> returns 0 results once I put it *alone* into a sub-query. Some
> issue about binding variables from full-text search?
>
>
>> Hm, I can't recall. Can you show the exact query so I can try to
>> reproduce?
Just tried to reproduce ... and I can't. Uuups. Probably I messed
something up before, sorry for the false alarm!! ;)

Should I ever come across it again, I'll let you know.


Cheers,
Marcel


>
>> Thanks, Pavel
>
>
>
>
> Cheers, Marcel
>
>
>
> On 02/08/16 15:18, Pavel Klinov wrote:
>> Hi Marcel,
>
>> I think using wildcards turns Lucene queries into constant-score
>> queries so by using wildcards you essentially throw away the
>> ranking. I'll need to double check though (I'm no Lucene
>> expert).
>
>> The main performance bottleneck in such queries is that FILTERs
>> with predicates such as STRSTARTS are applied to a lot of
>> literals. If you can use a full-text predicate in the same query
>> which would cut down the number of literals, e.g. to those
>> having 'Ben' somewhere, then applying the FILTER on top of that
>> should be faster (as well as ordering the results). The Stardog
>> query optimizer should first apply the search predicate and then
>> FILTERs in such cases. So the trick is to use a full-text
>> predicate which will match a superset of what you need.
>
>> Can't say much about custom scoring at this point but we can
>> look at that in the future.
>
>> Best, Pavel
>
>> On Tue, Aug 2, 2016 at 6:53 AM, Marcel Karnstedt-Hulpus
>> <marcel.k...@springernature.com
> <mailto:marcel.k...@springernature.com>
>> <mailto:marcel.k...@springernature.com
>> <mailto:mayurb...@gmail.com <mailto:mayurb...@gmail.com>>
> <mailto:mayurb...@gmail.com <mailto:mayurb...@gmail.com>>> wrote:
>>>>
>>>> Hello Pavel,
>>>>
>>>> The link you shared was a question asked by one of my
>>>> colleague. We do not have the option of having full text
>>>> search. We need to show results as user types in words,
>>>> hence we require that partial matches should be displayed.
>>>>
>>>> Thank you, Mayur
>>>>
>>>> On Monday, August 1, 2016 at 3:02:58 PM UTC+5:30, Pavel
>>>> Klinov wrote:
>>>>
>>>> It seems that your question is essentially a duplicate of
>>>> [1]. Please try the suggestion given there.
>>>>
>>>> Best, Pavel
>>>>
>>>> [1]
>>>>
>> https://groups.google.com/a/clarkparsia.com/d/msg/stardog/wxMSKxQ4rbs/UjPWL4-rBAAJ
>>>>
>>>>
>
>>
>
> <https://groups.google.com/a/clarkparsia.com/d/msg/stardog/wxMSKxQ4rbs/UjPWL4-rBAAJ>
>>>>
>>>>
>
On Mon, Aug 1, 2016 at 11:05 AM, Mayur Barge
>>>> <mayurb...@gmail.com <mailto:mayurb...@gmail.com>
> <mailto:sta...@clarkparsia.com <mailto:sta...@clarkparsia.com>> To
>> unsubscribe from this group, send email
>>>> to stardog+u...@clarkparsia.com
>>>> <mailto:stardog%2Bu...@clarkparsia.com>
>> <mailto:stardog%2Bu...@clarkparsia.com
> <mailto:stardog%252Bu...@clarkparsia.com>> For more options, visit
>> this
>>>> group at
>>>> http://groups.google.com/a/clarkparsia.com/group/stardog?hl=en
>>>>
>>>>
<http://groups.google.com/a/clarkparsia.com/group/stardog?hl=en>
>>>>
>>>>
>>>>
>>>>
> -- -- -- You received this message because you are subscribed to
>>>> the C&P "Stardog" group. To post to this group, send email
>>>> to sta...@clarkparsia.com <mailto:sta...@clarkparsia.com>
> <mailto:sta...@clarkparsia.com <mailto:sta...@clarkparsia.com>> To
>> unsubscribe from this group, send email
>>>> to stardog+u...@clarkparsia.com
>>>> <mailto:stardog%2Bu...@clarkparsia.com>
>> <mailto:stardog%2Bu...@clarkparsia.com
> <mailto:stardog%252Bu...@clarkparsia.com>> For more options, visit
>> this
>>>> group at
>>>> http://groups.google.com/a/clarkparsia.com/group/stardog?hl=en
>>>>
>>>>
<http://groups.google.com/a/clarkparsia.com/group/stardog?hl=en>
>>>>
>>>>
>>>>
>>>>
> -- -- -- You received this message because you are subscribed to
>>>> the C&P "Stardog" group. To post to this group, send email
>>>> to sta...@clarkparsia.com <mailto:sta...@clarkparsia.com>
> <mailto:sta...@clarkparsia.com <mailto:sta...@clarkparsia.com>>
>> <javascript:> To unsubscribe from this
>>>> group, send email to stardog+u...@clarkparsia.com
>>>> <mailto:stardog%2Bu...@clarkparsia.com>
>> <mailto:stardog%2Bu...@clarkparsia.com
> <mailto:stardog%252Bu...@clarkparsia.com>> <javascript:>
>>>> For more options, visit this group at
>>>> http://groups.google.com/a/clarkparsia.com/group/stardog?hl=en
>>>>
>>>>
<http://groups.google.com/a/clarkparsia.com/group/stardog?hl=en>
>>>>
>>>>
> --- You received this message because you are subscribed to the
>>>> Google Groups "Stardog" group. To unsubscribe from this
>>>> group and stop receiving emails from it, send an email to
>>>> stardog+u...@clarkparsia.com
>>>> <mailto:stardog%2Bu...@clarkparsia.com>
>> <mailto:stardog%2Bu...@clarkparsia.com
> <mailto:stardog%252Bu...@clarkparsia.com>> <javascript:>.
>
>>> -- -- -- You received this message because you are subscribed
>>> to the C&P "Stardog" group. To post to this group, send email
>>> to sta...@clarkparsia.com <mailto:sta...@clarkparsia.com>
> <mailto:sta...@clarkparsia.com <mailto:sta...@clarkparsia.com>>
> To
>> unsubscribe from this group, send email
>>> to stardog+u...@clarkparsia.com
> <mailto:stardog%2Bunsu...@clarkparsia.com>
>> <mailto:stardog%2Bunsu...@clarkparsia.com
> <mailto:stardog%252Buns...@clarkparsia.com>> For more
> options,
>> visit this
>>> group at
>>> http://groups.google.com/a/clarkparsia.com/group/stardog?hl=en
>>> --- You received this message because you are subscribed to
>>> the Google Groups "Stardog" group. To unsubscribe from this
>>> group and stop receiving emails from it, send an email to
>>> stardog+u...@clarkparsia.com
> <mailto:stardog%2Bunsu...@clarkparsia.com>
>> <mailto:stardog%2Bunsu...@clarkparsia.com
> <mailto:stardog%252Buns...@clarkparsia.com>>
>>> <mailto:stardog+u...@clarkparsia.com
> <mailto:stardog%2Bunsu...@clarkparsia.com>
>> <mailto:stardog%2Bunsu...@clarkparsia.com
> <mailto:stardog%252Buns...@clarkparsia.com>>>.
>
>
>> -- -- -- You received this message because you are subscribed to
>> the C&P "Stardog" group. To post to this group, send email to
>> sta...@clarkparsia.com <mailto:sta...@clarkparsia.com>
> <mailto:sta...@clarkparsia.com <mailto:sta...@clarkparsia.com>>
> To
>> unsubscribe from this group, send email to
>> stardog+u...@clarkparsia.com
> <mailto:stardog%2Bunsu...@clarkparsia.com>
>> <mailto:stardog%2Bunsu...@clarkparsia.com
> <mailto:stardog%252Buns...@clarkparsia.com>> For more
iQEcBAEBCgAGBQJXoNykAAoJELjmZAjIQprm1L8H/1Wrfx9LBas5TTkangmD/r6b
pTPHzFN6lMC32z/LRanv8nEofOiKJwv9sYqZucYczT4ZcmFhk9pzZZel2NrT2KRO
ttFba12J16pj0YJ4AkV//wwTOLiKi2NAMgD7+Um27Ipz+QyxUgqas28tfp/9sWOl
P/uO6aKq8HOWyAwDRjdK3X/hM3c1/obu3rKqnw+t7DhAbwrKU1NXHQQYwZ/KUH3S
lFktPyvt0Jb7mtkdzeVf4kEl62hmx2W7tHY07CjjZmMH3Jq9WEgE28sohA/zYQqg
KLwdt1VmOQo3L6ypuCEaNIfvY6ChOtda244+TAfNT5VCJdlxDh1/5sIr8VPwSV8=
=/O4T
-----END PGP SIGNATURE-----
Reply all
Reply to author
Forward
0 new messages