Re: [aseek-users] Ignoring the title, description, and keywords

Kir Kolyshkin <[email protected]> Fri, 08 Nov 2002 05:05:42 +0300
Newsgroups gmane.comp.web.aspseek.devel
Message-ID <[email protected]>
Matt Sullivan wrote:
> On Mon, 04 Nov 2002 at 02:54:42 +0300, Kir Kolyshkin wrote:
> 
> 
>>Francesco Caccavella wrote:
>>
>>>At 19.46 02/11/2002 -0500, you wrote:
>>>
>>>
>>>>Is there a way I can ask searchd to ignore the HTML title and the META
>>>>keywords and description when performing a search?
>>>>
>>>>I've configured my system to use tl=off, ds=off, and kw=off but that 
>>>>doesn't
>>>>seem to be working.
>>>>
>>>>Thanks!
>>>
>>>
>>>As I suggest between the lines some messages ago 
>>>(http://forum.aspseek.org/index.php?t=msg&goto=1472), in the next 
>>>release (when? ;-) would be useful to disallow (or simply diminish) the 
>>>weight of the meta keywords and description for the results (directly in 
>>>aspseek.conf or searchd.conf). Google ignores this meta too.
>>
>>Probably a better idea would be to implement two "custom"
>>meta-tags that could be used instead of "Keywords" and
>>"Description" by users which wants them, and have an additional
>>option in either searchd.conf or s.htm's "variables" section
>>to specify the defaults about which fields are to be used.
>>Matt, what's your opinion?
> 
> 
> Yes, I think there would be some benefit in this.
> 
> As I see it, this comes back to the "href text" modifications.  I had a
> working version of this some months back but it got put on hold while I
> re-wrote libaspseek and the PHP module and I haven't come back to it as
> yet.  I think that was at version 1.2.6 so I have some work to do to bring
> the patch up to date.
> 
> I had also started to implement domain name words using a slightly
> modified ternary tree of dictionary words to decompose the domain string
> into words.
> 
> Currently there is no space to support custom tags within the word indexes
> (all combinations of the 2 bits assigned for this are used) however with
> the href text modifications this will be extended to 4 bits of which the
> msb will indicate "Body" leaving 3 usable bits.  Out of this will be
> assigned "Title", "Description", "Keywords", "HREF Text" and "Domain Words"
> leaving 3 unused slots.

What about having a variable length of "section" part in splitting
into "section" and "word_position"? I mean, currently we have 2 bits for 
"section", your approach is to have 4 bits.

Taking into account the thing that we need to store large numbers into 
"word_position" only for the BODY section (this is the _main_ idea and
assumption), my proposal is to have it this way:

If the first bit (MSB) is 0, this means BODY, and only this one bit is
used for "section" (thus we are adding one extra bit to "word_position"
for the BODY section, which makes some good sense to me).

Otherwise (MSB is 1) the layout can be the following:
6 bytes for "section" (first is always 1, so we can have
2^5 = 32 sections):
100000 - TITLE
100001 - KEYWORDS
100010 - DESCRIPTION
100011 - HREF_TEXT
100100 - URL_TEXT (or "Domain Words")

10xxxx - RESERVED (11 combinations, probably can be used for, say,
          MP3's ID3 tags (Author/Title/Album/Year/Genre/Whatever),
          or some other cool things like image search
          (IMG ALT/TITLE/URL_TEXT/HREF_TEXT/SURROUNDING_TEXT)

11xxxx - CUSTOM DEFINED SECTIONS (up to 16). Actually people
          asked me a several times to have such an ability to
          search, say, for Author META, so we can at least
          reserve a space for it.

Matt, what do you think? To my mind, it looks quite OK, but code
changes needed can be quite massive (but the same is truth for your
4-bit proposal).

PS I'm still ill (and probably on some antibiotic drugs, too), so 
forgive (and fix) me if I am mistaken somewhere in the above mumblings.

> I agree that the order of relevance and the overall weight of these should
> be configurable.
> 
> I'd like to get the href text additions into the devel tree but I'm not
> sure of a timeframe for this yet.
> 
> I'm cross posting this to aseek-devel since I think the thread should be
> continued there.
> 
> 
> Matt.
>