Thread: Html parsing and inline elements
Hi everyone,
I was here wondering whether HTML parsing should separate tokens that are not separated by spaces in the original text, but are separated by an inline element. Let me show you an example:
SELECT to_tsvector('english', 'Hello<p>neighbor</p>, you are <strong>n</strong>i<em>ce</em>')
Results: "'ce':7 'hello':1 'n':5 'neighbor':2"
"Hello" and "neighbor" should really be separated, because <p> is a block element, but "nice" should be a single word there, since there is no visual separation when rendered (<em> and <strong> are inline elements).
Sorry if this has been asked before, but I couldn't find it anywhere.
Thanks in advance,
Marcelo.
Marcelo Zabani <mzabani@gmail.com> writes:
> I was here wondering whether HTML parsing should separate tokens that are
> not separated by spaces in the original text, but are separated by an
> inline element. Let me show you an example:
> *SELECT to_tsvector('english', 'Hello<p>neighbor</p>, you are
> <strong>n</strong>i<em>ce</em>')*
> *Results:** "'ce':7 'hello':1 'n':5 'neighbor':2"*
> "Hello" and "neighbor" should really be separated, because *<p>* is a block
> element, but "nice" should be a single word there, since there is no visual
> separation when rendered (*<em>* and *<strong>* are inline elements).
I can't imagine that we want to_tsvector to know that much about HTML.
It doesn't, really, even have license to assume that its input *is*
HTML. So even if you see things that look like <foo> and </foo> in the
string, it could easily be XML or SGML or some other SGML-like markup
format with different semantics for the markup keywords.
Perhaps it'd be sane to do something like this as long as the
HTML-specific behavior was broken out into a separate function.
(Or maybe it could be done within to_tsvector as a separate parser
or separate dictionary?) But I don't think it should be part of
the default behavior.
regards, tom lane
Hi, Tom,
You're right, I don't think one can argue that the default parser should know HTML.
How about your suggestion of there being an HTML parser, is it feasible? I ask this because I think that a lot of people store HTML documents these days, and although there probably aren't lots of HTML with words written along multiple inline elements, it would certainly be nice to have a proper parser for these use cases.
What do you think?
On Wed, Apr 13, 2016 at 11:09 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
Marcelo Zabani <mzabani@gmail.com> writes:
> I was here wondering whether HTML parsing should separate tokens that are
> not separated by spaces in the original text, but are separated by an
> inline element. Let me show you an example:
> *SELECT to_tsvector('english', 'Hello<p>neighbor</p>, you are
> <strong>n</strong>i<em>ce</em>')*
> *Results:** "'ce':7 'hello':1 'n':5 'neighbor':2"*
> "Hello" and "neighbor" should really be separated, because *<p>* is a block
> element, but "nice" should be a single word there, since there is no visual
> separation when rendered (*<em>* and *<strong>* are inline elements).
I can't imagine that we want to_tsvector to know that much about HTML.
It doesn't, really, even have license to assume that its input *is*
HTML. So even if you see things that look like <foo> and </foo> in the
string, it could easily be XML or SGML or some other SGML-like markup
format with different semantics for the markup keywords.
Perhaps it'd be sane to do something like this as long as the
HTML-specific behavior was broken out into a separate function.
(Or maybe it could be done within to_tsvector as a separate parser
or separate dictionary?) But I don't think it should be part of
the default behavior.
regards, tom lane
On Wed, Apr 13, 2016 at 12:57:19PM -0300, Marcelo Zabani wrote: > Hi, Tom, > > You're right, I don't think one can argue that the default parser should know > HTML. > How about your suggestion of there being an HTML parser, is it feasible? I ask > this because I think that a lot of people store HTML documents these days, and > although there probably aren't lots of HTML with words written along multiple > inline elements, it would certainly be nice to have a proper parser for these > use cases. > > What do you think? It sounds useful. -- 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 Wed, Apr 13, 2016 at 12:57:19PM -0300, Marcelo Zabani wrote:
> Hi, Tom,
>
> You're right, I don't think one can argue that the default parser should know
> HTML.
> How about your suggestion of there being an HTML parser, is it feasible? I ask
> this because I think that a lot of people store HTML documents these days, and
> although there probably aren't lots of HTML with words written along multiple
> inline elements, it would certainly be nice to have a proper parser for these
> use cases.
>
> What do you think?
It sounds useful.
It sounds like an external project/extension...
David J.
On Wed, Apr 13, 2016 at 6:57 PM, Marcelo Zabani <mzabani@gmail.com> wrote:
Hi, Tom,You're right, I don't think one can argue that the default parser should know HTML.How about your suggestion of there being an HTML parser, is it feasible? I ask this because I think that a lot of people store HTML documents these days, and although there probably aren't lots of HTML with words written along multiple inline elements, it would certainly be nice to have a proper parser for these use cases.What do you think?
I think it could be useful separate parser. But the problem is how to fully utilize it to facilitate ranking, for example, words in title could be considered more important than in the body, etc. Currently, setweight() functions provides this separately from parser.
Parser outputs tokid and token:
select * from ts_parse('default','<title>parser</title><body>text</body>');
tokid | token
-------+----------
13 | <title>
1 | parser
13 | </title>
13 | <body>
1 | text
13 | </body>
(6 rows)
Parser outputs tokid and token:
select * from ts_parse('default','<title>parser</title><body>text</body>');
tokid | token
-------+----------
13 | <title>
1 | parser
13 | </title>
13 | <body>
1 | text
13 | </body>
(6 rows)
If we change parser to output also rank flag, then we could use it to assign different weights.
On Wed, Apr 13, 2016 at 11:09 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:Marcelo Zabani <mzabani@gmail.com> writes:
> I was here wondering whether HTML parsing should separate tokens that are
> not separated by spaces in the original text, but are separated by an
> inline element. Let me show you an example:
> *SELECT to_tsvector('english', 'Hello<p>neighbor</p>, you are
> <strong>n</strong>i<em>ce</em>')*
> *Results:** "'ce':7 'hello':1 'n':5 'neighbor':2"*
> "Hello" and "neighbor" should really be separated, because *<p>* is a block
> element, but "nice" should be a single word there, since there is no visual
> separation when rendered (*<em>* and *<strong>* are inline elements).
I can't imagine that we want to_tsvector to know that much about HTML.
It doesn't, really, even have license to assume that its input *is*
HTML. So even if you see things that look like <foo> and </foo> in the
string, it could easily be XML or SGML or some other SGML-like markup
format with different semantics for the markup keywords.
Perhaps it'd be sane to do something like this as long as the
HTML-specific behavior was broken out into a separate function.
(Or maybe it could be done within to_tsvector as a separate parser
or separate dictionary?) But I don't think it should be part of
the default behavior.
regards, tom lane
On Wed, Apr 13, 2016 at 9:57 AM, Marcelo Zabani <mzabani@gmail.com> wrote:
Hi, Tom,You're right, I don't think one can argue that the default parser should know HTML.How about your suggestion of there being an HTML parser, is it feasible? I ask this because I think that a lot of people store HTML documents these days, and although there probably aren't lots of HTML with words written along multiple inline elements, it would certainly be nice to have a proper parser for these use cases.What do you think?
I recommend using Apache Tika [1] for plain text extraction from HTML. There are so many weird edge cases when parsing HTML that it is easier to use something that is already mature than reinventing the wheel.
Thanks,
Ryan Pedela