Why not keeping positions in GIN?

On 5/25/07, Hitoshi Harada <hitoshi_harada@forcia.com> wrote:

It's not only about Japanese. When you search "phrase" for text in English,
the same logic above will be needed. I don't research about tsearch2 but is
there any problem?? Also, in some case int-array inverted index needs the
entry positions as well, I guess. Obtaining positions with posting lists is
"general" enough for GIN, isn't it?

Is there any future plan around it?

We talked of this with Oleg and Teodor when I worked on GIN for
pg_trgm. I know there is a long term plan to solve this issue (and
especially improve ranking in full text search).

I'm not sure the position is general enough. What I'd like to have is
the ability to add metadata. For example, in the case of pg_trgm, I'd
like to have the length of the original string as it's a strong factor
in similarity calculation. Currently, we get a lot of results which
are rechecked after the first index pass: it's not very efficient.

--
Guillaume

On Fri, 25 May 2007, Hitoshi Harada wrote:

Hi,

I was walking through GIN am source code these days, and found that it has
only posting lists but no positions related those.

The reason I was doing that is, to try to implement n-gram text search index
on GIN for myself. As you know Japanese is not like English or other
European languages. If you write Japanese (or other 'not separated') text
index by n-gram, it should have entry positions on the entry as well as the
posting lists, because you must know if each split query key are joined with
each other in the data. To know this, position must be there.

FYI, Tatsuo uses tsearch2 for indexing japanese documents. But I agree,
n-gram index would be more universal for asian languages.

It's not only about Japanese. When you search "phrase" for text in English,
the same logic above will be needed. I don't research about tsearch2 but is
there any problem?? Also, in some case int-array inverted index needs the
entry positions as well, I guess. Obtaining positions with posting lists is
"general" enough for GIN, isn't it?

Is there any future plan around it?

Yes, we do have plans. See our todo, http://www.sai.msu.su/~megera/wiki/todo
You may read also FTSBOOK, http://www.sai.msu.su/~megera/postgres/fts/doc
and slides from PGCon2007, http://www.sai.msu.su/~megera/postgres/talks/fts-pgcon2007.pdf

Regards,

Hitoshi Harada

---------------------------(end of broadcast)---------------------------
TIP 4: Have you searched our list archives?

http://archives.postgresql.org

Regards,
Oleg
_____________________________________________________________
Oleg Bartunov, Research Scientist, Head of AstroNet (www.astronet.ru),
Sternberg Astronomical Institute, Moscow University, Russia
Internet: oleg@sai.msu.su, http://www.sai.msu.su/~megera/
phone: +007(495)939-16-83, +007(495)939-23-83

FYI, Tatsuo uses tsearch2 for indexing japanese documents. But I agree,
n-gram index would be more universal for asian languages.

Yeah, I know, but in tsearch2 for japanese sample you must use external
morphological analysis libraries to separate words. It is powerful but I
need more "lightweight" approach. Also especially when you search for
non-document(such like titles, names, or pattern in the genome), the
approach above is not so useful.

As I mentioned, GIN is also powerful for array data type search, so I am
very expecting it will have additional information.

Anyway, thanks a lot for much information. I try to read it.

Regards,

Hitoshi Harada

-----Original Message-----
From: Oleg Bartunov [mailto:oleg@sai.msu.su]
Sent: Saturday, May 26, 2007 10:12 PM
To: Hitoshi Harada
Cc: pgsql-hackers@postgresql.org
Subject: Re: [HACKERS] Why not keeping positions in GIN?

On Fri, 25 May 2007, Hitoshi Harada wrote:

Hi,

I was walking through GIN am source code these days, and found that it

has

only posting lists but no positions related those.

The reason I was doing that is, to try to implement n-gram text search

index

on GIN for myself. As you know Japanese is not like English or other
European languages. If you write Japanese (or other 'not separated')

text

index by n-gram, it should have entry positions on the entry as well as

the

posting lists, because you must know if each split query key are joined

with

each other in the data. To know this, position must be there.

FYI, Tatsuo uses tsearch2 for indexing japanese documents. But I agree,
n-gram index would be more universal for asian languages.

It's not only about Japanese. When you search "phrase" for text in

English,

the same logic above will be needed. I don't research about tsearch2 but

is

there any problem?? Also, in some case int-array inverted index needs

the

entry positions as well, I guess. Obtaining positions with posting lists

is

"general" enough for GIN, isn't it?

Is there any future plan around it?

Yes, we do have plans. See our todo,

http://www.sai.msu.su/~megera/wiki/todo

Hitoshi,

there is no problem to write n-gram dictionary for tsearch2 ! The problem
is in how to define word boundary.

Oleg

On Sat, 26 May 2007, Hitoshi Harada wrote:

FYI, Tatsuo uses tsearch2 for indexing japanese documents. But I agree,
n-gram index would be more universal for asian languages.

Yeah, I know, but in tsearch2 for japanese sample you must use external
morphological analysis libraries to separate words. It is powerful but I
need more "lightweight" approach. Also especially when you search for
non-document(such like titles, names, or pattern in the genome), the
approach above is not so useful.

As I mentioned, GIN is also powerful for array data type search, so I am
very expecting it will have additional information.

Anyway, thanks a lot for much information. I try to read it.

Regards,

Hitoshi Harada

-----Original Message-----
From: Oleg Bartunov [mailto:oleg@sai.msu.su]
Sent: Saturday, May 26, 2007 10:12 PM
To: Hitoshi Harada
Cc: pgsql-hackers@postgresql.org
Subject: Re: [HACKERS] Why not keeping positions in GIN?

On Fri, 25 May 2007, Hitoshi Harada wrote:

Hi,

I was walking through GIN am source code these days, and found that it

has

only posting lists but no positions related those.

The reason I was doing that is, to try to implement n-gram text search

index

on GIN for myself. As you know Japanese is not like English or other
European languages. If you write Japanese (or other 'not separated')

text

index by n-gram, it should have entry positions on the entry as well as

the

posting lists, because you must know if each split query key are joined

with

each other in the data. To know this, position must be there.

FYI, Tatsuo uses tsearch2 for indexing japanese documents. But I agree,
n-gram index would be more universal for asian languages.

It's not only about Japanese. When you search "phrase" for text in

English,

the same logic above will be needed. I don't research about tsearch2 but

is

there any problem?? Also, in some case int-array inverted index needs

the

entry positions as well, I guess. Obtaining positions with posting lists

is

"general" enough for GIN, isn't it?

Is there any future plan around it?

Yes, we do have plans. See our todo,

http://www.sai.msu.su/~megera/wiki/todo

You may read also FTSBOOK, http://www.sai.msu.su/~megera/postgres/fts/doc
and slides from PGCon2007,
http://www.sai.msu.su/~megera/postgres/talks/fts-pgcon2007.pdf

Regards,

Hitoshi Harada

---------------------------(end of broadcast)---------------------------
TIP 4: Have you searched our list archives?

http://archives.postgresql.org

Regards,
Oleg
_____________________________________________________________
Oleg Bartunov, Research Scientist, Head of AstroNet (www.astronet.ru),
Sternberg Astronomical Institute, Moscow University, Russia
Internet: oleg@sai.msu.su, http://www.sai.msu.su/~megera/
phone: +007(495)939-16-83, +007(495)939-23-83

Regards,
Oleg
_____________________________________________________________
Oleg Bartunov, Research Scientist, Head of AstroNet (www.astronet.ru),
Sternberg Astronomical Institute, Moscow University, Russia
Internet: oleg@sai.msu.su, http://www.sai.msu.su/~megera/
phone: +007(495)939-16-83, +007(495)939-23-83