What is going on with the SL Search ???

You're being rather unfair, Phil did mention that keyword repetition shouldn't work and as to how does keyword stuffing benefit the searcher, it doesn't, but it doesn't now anyway.

I have also mentioned that keyword repetition shouldn't work.
This is because it shouldn't.

It makes perfect sense for the GSA to positively take note of rank manipulation like keyword repetition. It's an enterprise system. The information function within an enterprise will wish to arrange things such that some information resources get priority. It's their search engine for their information.
That does not apply in SL.

If the GSA or whatever other search mechanism LL might use in the future is stuck with giving keyword repetition an advantage, then the response has to be to make keyword stuffing a TOS offence. LL are already clearly down the road on whacking "gaming".


Permitting an invisible area for extra text won't do anything to improve parcel presentation. It just gives a place where other gamers can't see part of your gaming.
If keyword stuffing on the visible page brings a ranking advantage, people are going to use it.
If they get an extra area to stuff. They'll game there as well.

Ban gaming.
Don't facilitate methods of hiding it.

Of course it will improve parcel presentation. If the description field had absolutely no search ranking then it would be used as a proper description.

You're missing a few points, Sling.

One is that the GSA is an internal system and does not expect to be spammed, so there is nothing in the way of spam checking in it, which means that spam works.

Another is that the idea is *only* about improving what users see - nice informative pages. It is nothing to do with getting rid of spam.

Another one is that the idea would not need any policing of any kind. Your alternative is for LL to make some thing s a ToS violation, which would need policing. And we all know what that means - hardly any anything gets policed.

As I said, if you have a better idea, let's hear it. Your idea of making things ToS violations can't work.

No. Not "as well". You seem to think that both the visible and invisible pages would affect the rankings, but that's not what I said. There would be absolutely no point in stuffing the visible page because it would have *no* effect on the rankings - the GSA would never see it. So, instead of parcel owners using that page for spam, they would make it as good as they could for users - to attract them to their places.

Why should you consider it a penalisation any differently from the penalising of searchers that happens widely, for example by separating the words "sex" and "bed"? If I am looking for "sex" then I don't want a furniture store. If I am looking for a bed I don't want listings of escorts and brothels. If I want a sexbed I will type in sexbed as one word.

Pep (If effective keywords were restricted to limited categories it would benefit more consumers than it would discommode, IMO)

Regardless, it can't be done with the GSA.

"Can't"?

Pep (It may not operate like that currently, but the mechanisms are now in place for discriminatory elimination to be effected.)

"bed"-sex

Explain please.

Nope. You have misunderstood.


This is the exactly point that I made, but in different words. The GSA is an enterprise system. There is no need to fight spam or any sort of deliberate ranking manipulation.
Rather, their is an advantage in allowing such deliperate manipulation so that the information function can mould what the enterprise sees.



Right now, we see parcel pages that are focussed more on gaming than on being informative.
What happens when such people get another area of text to game? They do the same there.
What happens if someone has a nice clean page but is being outranked by a key-word stuffing page? They start key-word stuffing, and their ranking rises. Everyone who currently stuffs will continue to stuff. Others will join in.
Nothing has changed.



The idea does nothing apart from hiding some of the gaming techniques.



Just adding more options for gaming won't be helpful to searchers either. That only helps the gamers.

On the other hand, making the various form of gaming TOS violations does help to put pressure on the gamers. You may recall forum discussions in which some people defended gaming on the basis that it wasn't against the TOS.

Banning something via TOS wouldn't magically change things overnight or even in the short term.
What it does do is to set down a standard of behaviour to be expected. It promotes a better culture.

People have pointed to other things that have been banned in a high-profile way, but are still to be found around us - e.g. Gambling, OTT advertising, predatory cutting.
These things are dying a slow death.


SO
LL can explicitly add things like keyword spamming to the TOS.
There is an alternative to an AR flood.




You want a "better idea". Here it is:

Run a second crawler, the function of which is to generate a list of parcels that are deemed to be keyword spamming.
Warn the perps
They persist? Remove from the index.
No 'Policing' - as in human eyeballs processing ARs.


That approach would allow LL to continue using the GSA, but remove the benefit of things like keyword spamming.
The fist step in that process is to make it clear in the TOS that such rank manipulation techniques are not to be used.

I understood a simple filter on search for "Adult words" is actually being implemented . A more sophisticated one is a trivial change.

Pep (The dam is cracked already; wait for the deluge)

I see your point.

The problem is that that the invisibility of the indexed page serves to mask manipulation exploits.
There would be a benefit in that the visible page would be cleaner if the content of that page were know to be totally ignored by the rankings.
*However, to get to see that cleaner page, the searcher is first presented with a list of parcels in which the ranking has been gamed.*

The major interest of the searcher has not been addressed.
What is being primarily addressed are the wants of parcel owners to manipulate their ranking without making their visible pages ugly. The issue of gaming is being swept under the carpet. This works against the interests of the searchers.


How would LL explain the use of the invisible page?
====================================
Q. What is this page for?
A. This is for manipulating the rankings in our search engine.
For a very simple example, if you sell dogs, Put dogs,dogs,dog,dogs,dogs. In there.
If another person selling dogs is outranking you, put more dogs in there.

And.. Don't tell anyone that we told you, but ... put anything you want in there. Nobody will ever be able to see what you doing, so put for instance "donkey, elephant" in there. Put lots of them. It will be undetectable. The page is never seen by anyone but you - and the automated indexing system.
People explicitly looking for elephants will visit your place and maybe buy your dogs. They are looking for animals are they not?
On the visible page, keep things a bit vague on the particular animals that you sell.

Also consider putting "free sex" in there a few times. It may pull eyeballs to your listing, and nobody will ever know. People don't spend ALL of their SL lookig for free sex. Sometimes they buy dogs.
===========================================

OK. The last bit is maybe a bit too chatty for the Knowledge Base, but if LL is going to facilitate stealth gaming, then they ow it to their users to makes things clear.

Yes but the cleaner description should give the searcher enough information as to whether their search has revealed content they are interested in.

For example if I search for furniture and see that the place is actually a free sex orgy room, the chances are it won't have what I'm looking for so I won't go there. There is little point in putting false keywords in there any more than there is now.

Search for prefabs and you'll see people don't care about keyword repitition, but there's not much information there for the person performing search, take that rubbish out of the eyeline.

Maybe even as a compromise list the parcel name tags and description tags much further down the page so the description and name are actually worthy and then check the tags but search is becoming a big old mess not because of false keywords but because of keyword repititions and the emphasis search places on keywords, it's one ugly looking set of results.

If the major problem with readability of pages is keyword repetition, then stop rewarding the behaviour.
The simplest solution would be an automated system whose sole mission is to identify gaming of pages. That does not require any changes to the GSA/whatever ranking system. It would run in parallel and removes offenders from Search.

No benefit in keyword-stuffing? Then - No keyword stuffing.


Benefits for searchers:
1. Cleaner pages.
2. Reduction in gaming of search.
Both are clear benefits to the searcher.

Benefits for SL:
1. The above benefits for the users
2. No need for Policing
3. No need for modifications to the parcel database
4. No need for development of the client and sim code to handle an extra text area.



Simply making the gaming undetectable primarily benefits the gamer.

Search will always be keyword based, it's not big enough not to be keyword based, people writing in plain English are wasting keyword space.

Making that keyword space a different area and giving a plain description field that isn't indexed for search purposes helps the searcher a hell of a lot.

If more space is required, then making the existing space larger would be the simplest solution, both for the developers and the parcel owners.
Given the amount of asset server space taken up by everyone's Inventory, a few k of text per parcel wouldn't be a noticeable load.

Even if a separate section were created for indexing 'only by keywords', this should
- be visible to all to detect misleading keywords.
- be crawled automatically to punish other abuse.

So I have an escort club (Escort Island if you care, in Cerura). There are two parcels, the main club, and a side room for webcam, cause people who look for cam are a different market than people who look for cyber.

The main parcel has traffic of 12,800, the cam parcel has traffic of 45. The main parcel has been there about 18 months longer (we bought the side parcel more recently). The main parcel has classifieds running, has way more profile picks and teleport landings. They have substantially the same words in the title and description. So in every way that is supposed to count, the main parcel should rank better,

And yet....the cam parcel scores 10 places higher under Search/All for escort (25th vs 35th).

SO the question is, what are they using now to rank places?

For all I care... Linden adds a randomizer into the search engine, making the results look different on a daily basis.
Sure, some stores have nicer stuff than others. But too many people start looking alike, living in the exact same kind of homes, have the exact same poses built into their beds. A bit of variety won't hurt anyone, and there's actually some very good stuff available from rather unknown sources, so that would be a help too.

Currently, search isn't anything but a camping contest with bonus points to those with expensive adds. There often isn't much of a connection between search and find, apart from people throwing around random keywords everywhere.

It's quite interesting.
The Webcam parcel is actually at position 18 right now.

The long main parcel page is a tour-de-force in repetition of "escort".
The very short webcam page is almost mute by comparison of the keyword repetition count.
However, looking at both pages side by side, there is something very different about them - and it's not the length.

Here's something.

If I were writing an algorithm to rank documents in an enterprise, I'd consider this.
- Most well written documents begin with a short overview - a management summary. I might give more weight to words in the body text at the beginning of a document than I would to words in body text further down.

Your webcam parcel is "escort" oriented in the list of objects.
Your main parcel doesn't mention "escort" in that list until the 12th object




Yes. it's crazy, maybe, but who the hell knows.
Just for a laugh, see what if any the effect of changing those object descriptions would be.
Eliminate the possibility.

A longer description would just lead to more keywords being used.



I'm not averse to that, stick them at the bottom.

How about something radically simple, at least for shopping sites:

Limit the search index to the descriptions of the objects that are for sale. Limit the description to some moderate length, say 60 characters. Remove special characters from the description, don't count duplicate words. Each indexed object has a minimum cost of one prim. Nothing else counts. If you feel that's not enough for your needs, buy a classified.

This will hurt people who sell mulitple items through a single vendor. Sorry, that will have to wait till V2 - perhaps with a standard vendor API, so that the engine scanning for objects can cycle through all available items.

Put smarts into the search engine to recognize related words, so that searching for "bed" and "beds" are considered identical.

If I'm searching for something to buy, then the only thing that's really relevant is whether or not such an object is for sale. The parcel description is irrelevant. Keywords are irrelevant. Either you have it for sale or your don't.

Items are listed alphabetically, and the couches are no mod and owned by someone else, so I cant change their names, but I changed the others. Have to wait a bit to see what happens.

Sling. I'm not going to reply to your posts bit by bit, as I see no need to. But I'll respond to a few points you made.

1. Ok, make the "hidden" page viewable by request. It would allow people to see how others got above them but that's not too bad as long as the viewable, informative page is the one that is returned when clicking on a search result.

2. Your idea of just putting "dogs dogs dogs dogs" etc. in the hidden page doesn't work with the GSA. There is a bit more to it than that. And throwing in something like "free sex", or whatever it was you said, doesn't work either with the GSA. It's just not as simple as that.

3. Your idea of expanding one or more of the current spaces would just mean that the extra space(s) will be used for more ugliness for users to see.

4. Keyword stuffing can't be outlawed, and it can't be defeated as long as LL uses the GSA. How could a rule be made? Set a maximum percentage for any single word? People word simply use the maximum allowed. There are no rules that can be layed down that will prevent it. Making things a ToS violation doesn't work, as we can see all too well with microparcels and bots. All a ToS violation does is reduce something. LL never follows things through enough to eradicate anything. They are not that interested. With LL, the answer is never "make a rule, make a rule" because they never hold up their end when they are pushed into making rules. It's far better to make changes that work automatically than to make rules. Or, first make the change that works autromatically, to the benefit of users, and then press for rules to deal with the content of the hidden page. The users would be well served from the start and, if LL then makes a rule about the hidden page, what happens in them will reduce - not go away, but reduce.

5. As a user, I don't care how places reached the top if the results are relevant to my search. This is probably where we mostly part company on this idea. Your objective in the discussion seems to be getting rid of gaming, so any changes must be intended to reduce or get rid of gaming, irrespective of whether or not users get to see nice pages. The objective in my idea is to make the search better for users, by returning useful, informative pages, rather than the stuffed pages that users are presented with now, and that will only get worse. I'm not going to argue with you about it. We have different objectives, and so we see this idea differently, that's all.

As a totally separate exercise following that one, I'd be curious to see if the very high number of repetitive entries was a help or a hindrance.
There are all sorts of other factors involved, but the pages of other higher ranked parcels don't particularly indicate that it's a huge help.