Thread: 9.6 phrase search distance specification
Does anyone know why the phrase distance "<3>" was changed from "at most three tokens away" to "exactly three tokens away"? I looked at the thread at: https://www.postgresql.org/message-id/flat/33828354.WrrSMviC7Y%40abook and didn't see the answer. I assume if you are looking for "<3>" you would want "<2>" matches and "<1>" matches as well. -- Bruce Momjian <bruce@momjian.us> http://momjian.us EnterpriseDB http://enterprisedb.com + As you are, so once was I. As I am, so you will be. + + Ancient Roman grave inscription +
Bruce Momjian <bruce@momjian.us> writes: > Does anyone know why the phrase distance "<3>" was changed from "at most > three tokens away" to "exactly three tokens away"? So that it would correctly support phraseto_tsquery's use of the operator to represent omitted words (stopwords) in a phrase. I think there's probably some use in also providing an operator that does "at most this many tokens away", but Oleg/Teodor were evidently less excited, because they didn't take the time to do it. The thread where this change was discussed is https://www.postgresql.org/message-id/flat/c19fcfec308e6ccd952cdde9e648b505%40mail.gmail.com see particularly https://www.postgresql.org/message-id/11252.1465422251%40sss.pgh.pa.us regards, tom lane
On Tue, Aug 9, 2016 at 01:58:25PM -0400, Tom Lane wrote: > Bruce Momjian <bruce@momjian.us> writes: > > Does anyone know why the phrase distance "<3>" was changed from "at most > > three tokens away" to "exactly three tokens away"? > > So that it would correctly support phraseto_tsquery's use of the operator > to represent omitted words (stopwords) in a phrase. > > I think there's probably some use in also providing an operator that does > "at most this many tokens away", but Oleg/Teodor were evidently less > excited, because they didn't take the time to do it. > > The thread where this change was discussed is > > https://www.postgresql.org/message-id/flat/c19fcfec308e6ccd952cdde9e648b505%40mail.gmail.com > > see particularly > > https://www.postgresql.org/message-id/11252.1465422251%40sss.pgh.pa.us Ah, I know it was discussed somewhere. Thanks, the phraseto_tsquery tie-in was what I forgot. -- Bruce Momjian <bruce@momjian.us> http://momjian.us EnterpriseDB http://enterprisedb.com + As you are, so once was I. As I am, so you will be. + + Ancient Roman grave inscription +
On Tue, Aug 9, 2016 at 11:58 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
Bruce Momjian <bruce@momjian.us> writes:
> Does anyone know why the phrase distance "<3>" was changed from "at most
> three tokens away" to "exactly three tokens away"?
So that it would correctly support phraseto_tsquery's use of the operator
to represent omitted words (stopwords) in a phrase.
I think there's probably some use in also providing an operator that does
"at most this many tokens away", but Oleg/Teodor were evidently less
excited, because they didn't take the time to do it.
The thread where this change was discussed is
https://www.postgresql.org/message-id/flat/ c19fcfec308e6ccd952cdde9e648b5 05%40mail.gmail.com
see particularly
https://www.postgresql.org/message-id/11252.1465422251% 40sss.pgh.pa.us
I would say that it is worth it to have a "phrase slop" operator (Apache Lucene terminology). Proximity search is extremely useful for improving relevance and phrase slop is one of the tools to achieve that.
On Tue, Aug 9, 2016 at 12:59 PM, Ryan Pedela <rpedela@datalanche.com> wrote:
On Tue, Aug 9, 2016 at 11:58 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:Bruce Momjian <bruce@momjian.us> writes:
> Does anyone know why the phrase distance "<3>" was changed from "at most
> three tokens away" to "exactly three tokens away"?
So that it would correctly support phraseto_tsquery's use of the operator
to represent omitted words (stopwords) in a phrase.
I think there's probably some use in also providing an operator that does
"at most this many tokens away", but Oleg/Teodor were evidently less
excited, because they didn't take the time to do it.
The thread where this change was discussed is
https://www.postgresql.org/message-id/flat/c19fcfec308e6ccd9 52cdde9e648b505%40mail.gmail. com
see particularly
https://www.postgresql.org/message-id/11252.1465422251%40sss .pgh.pa.us I would say that it is worth it to have a "phrase slop" operator (Apache Lucene terminology). Proximity search is extremely useful for improving relevance and phrase slop is one of the tools to achieve that.
Sorry for the position of my signature....
Ryan
On Tue, Aug 9, 2016 at 9:59 PM, Ryan Pedela <rpedela@datalanche.com> wrote: > > > I would say that it is worth it to have a "phrase slop" operator (Apache > Lucene terminology). Proximity search is extremely useful for improving > relevance and phrase slop is one of the tools to achieve that. > It'd be great if you explain what is "phrase slop". I assume it's not about search, but about relevance.
On Thu, Aug 11, 2016 at 9:27 AM, Oleg Bartunov <obartunov@gmail.com> wrote:
On Tue, Aug 9, 2016 at 9:59 PM, Ryan Pedela <rpedela@datalanche.com> wrote:
>
>
> I would say that it is worth it to have a "phrase slop" operator (Apache
> Lucene terminology). Proximity search is extremely useful for improving
> relevance and phrase slop is one of the tools to achieve that.
>
It'd be great if you explain what is "phrase slop". I assume it's not
about search, but about relevance.
Sure. An exact phrase query has slop = 0 which means find all terms in the exact positions relative to each other. Phrase query with slop > 0 means find all terms within <slop> positions relative to each other. If slop = 10, find all terms within 10 positions of each other. Here is a concrete example from my current work searching SEC filings.
Bill Gates' full legal name is William H. Gates, III. In the SEC database [1], his name is GATES WILLIAM H III. If you are searching the records of people within the SEC database and you want to find Bill Gates, most users will type "bill gates". Since there are many people with the first name Bill (William) and the last name Gates, Bill Gates most likely won't be the first result with a standard keyword query. Likewise an exact phrase query (slop = 0) will not find him either because the first and last names are transposed. What you need is a phrase query with a slop = 2 which will match "William Gates", "William H Gates", "Gates William", etc. There is still the issue of Bill vs William, but that can be solved with synonyms and is a different topic.
Thanks,
Ryan
On Thu, Aug 11, 2016 at 10:42 AM, Ryan Pedela <rpedela@datalanche.com> wrote:
One more thing. In that trivial example, an AND query would probably do a great job too. However if you are searching for Bill Gates in large text documents rather than a list of names, an AND query will not give you very good results because the words "bill" and "gates" are so common.On Thu, Aug 11, 2016 at 9:27 AM, Oleg Bartunov <obartunov@gmail.com> wrote:On Tue, Aug 9, 2016 at 9:59 PM, Ryan Pedela <rpedela@datalanche.com> wrote:
>
>
> I would say that it is worth it to have a "phrase slop" operator (Apache
> Lucene terminology). Proximity search is extremely useful for improving
> relevance and phrase slop is one of the tools to achieve that.
>
It'd be great if you explain what is "phrase slop". I assume it's not
about search, but about relevance.Sure. An exact phrase query has slop = 0 which means find all terms in the exact positions relative to each other. Phrase query with slop > 0 means find all terms within <slop> positions relative to each other. If slop = 10, find all terms within 10 positions of each other. Here is a concrete example from my current work searching SEC filings.Bill Gates' full legal name is William H. Gates, III. In the SEC database [1], his name is GATES WILLIAM H III. If you are searching the records of people within the SEC database and you want to find Bill Gates, most users will type "bill gates". Since there are many people with the first name Bill (William) and the last name Gates, Bill Gates most likely won't be the first result with a standard keyword query. Likewise an exact phrase query (slop = 0) will not find him either because the first and last names are transposed. What you need is a phrase query with a slop = 2 which will match "William Gates", "William H Gates", "Gates William", etc. There is still the issue of Bill vs William, but that can be solved with synonyms and is a different topic.