I’ve been thinking a lot lately about Second Life search, and the complaints that are made about it. That most say that it isn’t as good as Google Web-search, that some complain that the new-new system has them ranked lower down than the old-new system did, that they can’t figure out how to boost their rankings and so on.
I’ll start off with two maxims:
1. The goal of a search-engine is to produce search results ranked by relevancy.
2. The goal of SEO (so-called Search-Engine Optimisation) is to reduce or eliminate relevancy in search results.
Now, before you choke on your coffee and attack me with forks, think about these two items for a moment. The hallmark of a successful search-engine is relevancy… the quality of results, that is, the likelihood that you’re going to find exactly what you wanted in the first few results, given a sufficiently well-formed query.
The top Web search-engines do just that. That’s why they’re the top engines, and not the also-rans.
SEO, on the other hand, (as it is commonly practiced) is a grab-bag of methods to increase your ranking in search results relative to (say) your competitors, without necessarily increasing relevancy. Why shouldn’t you be number one? Unfortunately, everyone would like to be number one.
SEO is, essentially, the natural enemy of the search-engine. It is to the benefit of the search-engine operator, and to search-engine users that no third-party be able to influence their result-rankings without making a real change in relevancy. Because if they can, do you know what you get? You get the classifieds section in the newspaper, which you and I no longer read, because we’ve got high-relevancy search-engines.
Let’s look at a couple of the criticisms individually. Second Life search isn’t as good as Google Web-search. Well, of course it isn’t. Even when search was using Google’s software for the job (that was ‘old-new’ search, for reference), it didn’t work nearly as well as the same software does when it searches the Web. That’s because Google searches the Web. Google indexes text. Lots and lots and lots of text. Yummy, delicious text.
When it comes to indexing a Second Life land parcel, there’s very very little text to chew on. Hardly any, actually, and the small space for descriptions tended to lead to false-triggering of Google’s anti-cheat algorithms, pushing down the rankings of parcels that might otherwise have ranked higher.
Personally, I think every parcel should have gotten a page on the Second Life Wiki, allowing the parcel holder to write long, detailed descriptions. Google Search would have loved that, and would have been able to produce brilliantly relevant results from it, performing just as well at finding relevant results as Google does with any Web-pages. That would have been a small change for a huge gain in effectiveness. It just didn’t happen.
The next complaint is heard from those whose rankings have dropped under the latest search system (the ‘new-new’ search). When you’ve dropped from (say) third place to twenty-third place, it’s hard to find good things to say about the search system. Mind you, 20 other parcel-holders just went up one rank (so to speak) and seem to have no cause for complaint. This all has to do with how the search-engine defines relevancy and how much it has to chew on – which, as I mentioned, isn’t very much at all.
People complain that they don’t understand how parcels are ranked in the results, and so they cannot manipulate their ranking. I think we covered that when I was talking about SEO. If, of course, the algorithm was widely known, the first few pages of results would simply be irrelevant listings for people who want to be number one. And then we’re back to newspaper classifieds.
Q: How many SEO specialists does it take to change a lightbulb, light bulb, light, bulb, lamp, cheap lamp, low cost light bulb, night light, illumination products, fluorescent tube, outdoor lighting, low energy light, low energy lighting system, power saving lights, bathroom lights, illuminated range hoods, discounted lighting products, solar garden light?
A: All of them. Constantly.
I shall believe it is the fault of the limited data for a search engine to chew on when it allows me to use a “well-formed query”, and still gets the mess that the Linden Lab search systems produce.
As a working SEO Consultant in SL, I have a few (hundred) thoughts on what SEO means inside SL.
What you said about SEO is correct in the “Traditional” environment of the WWW. But when it comes to the content-deprived world of Land Parcels and their pathetic ability to show relevancy for what the land owner actually sells (or provides for consumption by the rest of SL), the purpose I’ve found most productive for my clients is to trick the Search Indexer to get factual and correct relevance for the parcel … AND to get the parcel to score (rank) higher than those parcels included in search results that have nothing whatsoever to do with the search terms provided.
It’s demoralizing when your parcel appears at #23 … sure enough. But when the 22 parcels above yours have nothing whatsoever to do with the search terms entered, and no amount of tweaking, packing or spamming can elevate your parcel above those other 22 non-relevant results, it’s downright maddening.
My Client list is chock full of folks that daily watch their sales wither away, while people complain daily that “Search got me nothing worthwhile”. Those that seem to be “winning” strip their parcels of everything, remove all objects from search, and strip their Land Name and Description to 2-5 words each. But how can they be “Relevant” when not a thing can be derived from their information.
Since the new-new Search was unveiled, and after their tweaks to improve relevancy, I find more and more empty parcels near the top of every search term … and those that actually ARE relevant pushed further and further down. But hey, it’s “More Relevant” by the magical formula that the SL Search Team has set so it must be okay.
Nevertheless, my feedback from my Clients and from their Customers is … “Bah on Search. I’ve quit using/depending on it and now use only word of mouth to find a store.” I’ve taken to counseling my Clients to “do these steps, it’s the best you can hope for in this current environment” and to just go back to the old traditional means of Marketing. And those steps? “Strip your objects from Search. Strip your Land Name to just 2-3 words that are your primary market niche and strip your Land Description to 2-3 short sentences with your Primary (and possibly secondary) search terms.”
It works. It gets them at least in the top 20 … and then I counsel them to just pray folks ignore the 20 or so above them that have no relevance whatsoever and wind up at their store anyway.
Frustrated? Yeah. But even more I am disappointed and sad. I’m watching businesses close, people shut down and leave SL … and their product niches not being replenished by “Fresh Blood” because there are no new people trying to enter the market.
We are watching the old die off and leave, the new find disappointment and roadblocks too cumbersome to surmount … and all in the name of achieving “Relevance” in the Search Engine. What SL (and LL) needs to realize is that the Search Engine SHOULD BE used to provide a list of those establishments that are active and vibrant merchants with a good product selection for people to visit. Not big stores, not old stores, not fancy stores … but actual STORES that sell what the customer actually searched for.
As you mention, there are many ways to give the Search Engine lots of good relevant text from which to derive true relevance. And the new Search Engine has access to information such as traffic, sales, etc. that can give so much more info. Instead LL has chosen to remove and remove and carve off even more, until what we have left is empty parcel pages, a total lack of useful (albeit “relevant”) results … and the customer base less and less confident in the Search Engine as a useful tool.
*Sigh* But hey … it’s Relevant … right? *facepalm*
I pretty much agree with Darrius. I disagree that LL tried to make the SL search about relevance. I think they tried to be ‘fair’. As is the case with most of those trying to be ‘fair’ they lower everyone to the same level of misery.
Now non-relevant sites are as likely to appear in results as relevant sites. How could one be more fair? (that is sarcasm)
Also, by purchasing advertising one is supposed to appear in search in appropriate places. Google’s AdWords works amazing well. LL’s ads draw a large number of complaints. Enough so that after asking around in the forum and in world I’ve decided to avoid paid SL advertising.
I do SEO work for several companies. I achieve positions for them by making their sites more relevant than their competitors sites. If SEO folks decide to work with Google rather than against them, Google provides lots of help.
Its all about getting the right people together. People wanting to buy a camera need to hook up with those wanting to sell a camera. That is both Google’s and the SEO’s goal. Just getting more people there is no help to the client. The client needs people that will buy.
For my AdSense campaigns CPM can encourage one to go for massive random visitors. But, when one is paying for clicks and trying to sell a service or product, specific visitors are needed. Massive random visitors just drives cost up without affecting sales. Good SEO will bring relevant visitors to the site.
One of the things I frequently see on Marketplace searches is some merchant using the same list of keywords for everything he sells, effectively a keyword set for the store rather than the individual item.
At least the marketplace has its category system, however vague it might be.
But you can see the effect when you use the “Communities” option. It’s just a pre-load of the search.
For my money, SEO has become a tainted term. There are some pretty scurrilous tricks used on the web, worse than all those “models” and “dancers” used to pump up traffic numbers in SL. Careful site design is one thing, but that should be as much for the human user as for the search engine robots.
The SL search space just isn’t going to respond to those sorts of trickery.
Search in SL never worked worse or better than now. Search on SLexchange only worked “better” than the one which is active on the SLmarketplace because there was not such a sheer flood of products as there is now.
I do not doubt that the search function could be done more effectively and that all the “promo” tricks could be fought more effectively. But I have no idea how. Maybe Linden Lab should return to their original and excessive enough mission, which is providing server space and keeping the Linden Dollar exchange going, making the basical server and client software somewhat usable – and leave website management and http://www. stuff to some http://www. specialists.
Vivienne. we’re all relying on the anecdotal data of our own experiences, but I can’t contradict you on that. Apart from the sheer volume of the search-space, I can’t see anything that’s really changed. And you can say the same about other aspects of Second Life.
And so the question has to be, why don’t we see improvements? (Please, no arguments about Mesh: as far as a non-creator is concerned, it isn’t so different from Sculpts.) There’s huge amounts of work going on behind the scenes, with barely any results.
My hypothesis is that the people who might have understood the big picture have mostly gone. Once the Viewer design was contracted out for Viewer 2, the long term ability to maintain and enhance the viewer depended on the documentation quality.
XStreet to Marketplace–is there some of the same problem there?
This talk about SEO and relevancy degeneration reminds about a phenomenon that has been spotted in many evolutionary algorithm programs; I couldn’t find the specific example i was trying to remember, but check out this excerpt from page 81 of the book “Evolvable Systems: From Biology to Hardware” ( http://books.google.com.br/books?id=wCDvP-uISb0C&pg=PA81 ) Google led me to:
“Evolutionary computation does it wrong.
That’s not to say that an evolutionary algorithm doesn’t evolve solutions. Using
principles of selection, reproduction with inheritance, and variability, we evolve good
solutions to many problems all the time. The trouble is, to make an evolutionary
algorithm evolve, you often need to be trained in the black arts of genetic
representation wrangling and fitness function fiddling.
Along the way you’ll get a few glimpses of the creativity of evolution through the
bugs in your code: the little loopholes that are ruthlessly exploited by evolution to
produce unwanted and invalid solutions. Music with notes inaudible to the human ear,
designs with components floating in mid-air, fraud detectors that detect everything
and yet find nothing useful, virtual creatures that sidestep inertia and flick themselves
along in bizarre ways. Each result fascinating, and each prevented by the addition of
another constraint by the developer. The bugs are never reported in any publication,
and yet they point to the true capabilities of evolution
…
Evolution is not an optimiser. It is a satisficer, exploiting every conceivable (and
inconceivable) way to make the minimum necessary adaptation. Evolution’s greatest
and unrivalled skill is its ability to find and use every avenue to achieve almost
nothing. In nature it is able to do this trick often enough to generate ingenuity and
complexity beyond our comprehension”
—
The search engine ranking is the fitness function, but instead of programs randomly twiddling parameters it’s humans adjusting the contents of pages. SEO techniques evolve thru trial and error often exploiting bugs and loopholes in the fitness function, the ranking algorithm. Search results are not ranked by actual relevancy, unfotunatly artificial intelligence is not there yet (or fortunatly if you believe the odds of machine uprise are high enough). Search engines don’t actually understand what the contents of each entry in their database is about, it’s just a bunch of numbers for them.
I think there are some points missing here, optimising your parcel for search isn’t in itself bad, if you name your products by funky names, you won’t help yourself getting found in search, if you’re selling a black leather jacket, put that in your item’s name or description, calling it some funky name alone isn’t good for the merchant or the person looking for a black leather jacket.
Doing the above is about making your items and parcels relevant, it’s not trying to reduce or eliminate relevancy, it’s about making your parcel and items relevant.
Where search is a let down is with regards to the paid classifieds bar, these results are sorted by price not relevancy, which means irrelevant results are likely to appear near the top.
I don’t think that new new search is that bad in terms of relevancy for places, but when lots of places are competing you will always have people wondering why parcel a rates higher than parcel b, this is an old old complaint, what has always infuriated people is when a seemingly irrelevant parcels ranks higher than relevant parcels.
new new search doesn’t support boolean operators, unlike old new search, which is a shame.
The new search works great for me. My top results are always relevant. My #1 search result is usually what I wanted. The difference between the new search and the old, original V2 one, is night and day.
It seems like before I’d search for a parcel name that I knew existed but had no landmark for and get a whole list of parcels that hijacked the keyword, as if keyword spamming meant more than actual parcel names. I’d also get dropped in locations where stores used to be, or satellite locations rather than the main location of stores.
Sometimes mishaps still happen, but its nothing at all like it was before. Before was absolutely terrible and I’ve noticed folks really aren’t complaining about search as much as they used to. I’d almost forgotten it was a common complaint since it’s been so reliable for me the last few months.