Marketplace Disastrous Consequences

[Sally Audebarn]

Your intention is good, but the fact remains that the search engine is broken. It simply is and no matter how many  "work arounds" you suggest (whether they be effective or not), the aforementoned fact will not be altered. Something must be done by LL  to facilitate the Marketplace's much needed repair. All the blame cannot be thrown at us merchants and doing so is blatantly unfair. 

[Pamela Galli]

Yep this is what happened when everything was migrated from Xstreet, except we were working with the devs, and they could see that the algorithm was ranking cheap crap highest, and needed further tweaking.

The current commerce team, of course, likes to work under cover of darkness. After YEARS of broken search, they have presented this new version of search as a fait accompli and are done with it. They have checked it off their list. 

What do they care if it looks like the SL Marketplace sells most old crap? 

[ChinRey]

---

Pamela Galli wrote:

... they have presented this new version of search as a fait accompli ...

---

You mean a fail accompli, don't you?

But to be fair, the Commerce Team are listening both to buyers and sellers these days. They are making progress and I would be very surprised if they don't change the search ranking algorithm very soon.

The problem is that the overall progress is way too slow! It took them nearly a year to come up with this change and there are so many other more important issues to fix. SL50B is just 37 years away and at this speed we won't have a presentable Marketplace ready by then. That's not even a joke.

[Adams Scarmon]

All these changes to the MP and they still haven't figured out a way to EXCLUDE demo's from searchings.

[Caren Jewell]

I read this Infos and thank you all for spending much time to clear out what happen and how can change something better

 

What I don't understand is that LL change the MP but take never a look back to change older Issues too. How long must I look at my old Items I delete 7 years ago and have them just right in my List again and again .. to sort this out make sense but it should be a big problem to fix old Issues. I have in my List tons of old Items that I delete tons of years ago - and at last, I can understand the new changes but I have problems with new changes if LL don't spend one minute to think about the consequences for all and changes on ways like "eat or die"

 

 I be willing to change my title and keywords - no problem - it will need tons of time for me, no problem. But only IF I know that LL comes not in short time and changes things again or fix or have new "nice" Ideas

[Codex Alpha]

---

Sally Audebarn wrote:

Your intention is good, but the fact remains that

the search engine is broken

. It simply is and no matter how many  "work arounds" you suggest (whether they be effective or not), the aforementoned fact will not be altered. Something must be done by LL  to facilitate the Marketplace's much needed repair. All the blame cannot be thrown at us merchants and doing so is blatantly unfair. 

---

I wasn't arguing whether the search engine is broken or not.

In my opinion, it is working as intended (and demonstrated how it does) in some areas, and in others I agreed there is something to look at.

In the meantime, fix your listings to adhere to the Marketplace Listing Guidelines (and everyone else who is currently complaining) so that we can get an accurate picture of what is really going on. They're not 'workarounds', they're requirements of you as a merchant to assist in the Marketplace working correctly.

I hope you and the others who quickly want to find fault in my argument realize this and act likewise.

Accept responsibility for how you list your items, and the effect it would have if and when the Marketplace were truly optimized. I only suggested why you and others WOULD NOT BE FOUND if the Marketplace WORKED AS I WAS SUPPOSED TO, and why you may not be found currently and in future searches.

 

---

Caren Jewell wrote:

I read this Infos and thank you all for spending much time to clear out what happen and how can change something better

 

What I don't understand is that LL change the MP but take never a look back to change older Issues too. How long must I look at my old Items I delete 7 years ago and have them just right in my List again and again .. to sort this out make sense but it should be a big problem to fix old Issues. I have in my List tons of old Items that I delete tons of years ago - and at last, I can understand the new changes but I have problems with new changes if LL don't spend one minute to think about the consequences for all and changes on ways like "eat or die"

 

 I be willing to change my title and keywords - no problem - it will need tons of time for me, no problem. But only IF I know that LL comes not in short time and changes things again or fix or have new "nice" Ideas

---

It might just be a simple case of - it's not their priority.

If a merchant doesn't like the results, they can always pack up their shop and leave...

[LuvlyDay]

Im all in for Upgrading and Improving. But obviously, this New Marketplace isnt going well for  most of us. Newbies have no idea what thier doing and how to fill the searches. They dont even know how to type  demo...to get a demo. They are simple people, they want easy, they dont want complicated crap. Who has time for this?  When i type in a name a big list appears which is crap i dont want, a huge distraction.  Most people that have been hear for years have had a HUGE DROP in sales. If we dont make money, Second Life doesnt make money. As a merchant that has been here for 8 years, at the moment just happy to get someone to buy a demo.

[Codex Alpha]

---

LuvlyDay wrote:

Im all in for Upgrading and Improving. But obviously, this New Marketplace isnt going well for  most of us. Newbies have no idea what thier doing and how to fill the searches. They dont even know how to type  demo...to get a demo. They are simple people, they want easy, they dont want complicated crap. Who has time for this?  When i type in a name a big list appears which is crap i dont want, a huge distraction.  Most people that have been hear for years have had a HUGE DROP in sales. If we dont make money, Second Life doesnt make money. As a merchant that has been here for 8 years, at the moment just happy to get someone to buy a demo.

---

I just don't see what the matter is. I'm finding anything I want.

Try entering these search terms;

"Diamond Wedding Ring"

"Medieval Fruit Stand"

"Red Polka Dot Dress"

"Chain Link Fence"

As I type, the Marketplace suggest several items that use the keywords I typed in their product listing titles. I would prefer it also seeks the actual keyword list (at product listing time), but since that is rampantly abused - I'm fine with the MP searching titles instead. If the product is not accurately described in it's title, then it shouldn't be considered.

I had no problem finding a 'demo' item in the search...

"demo green striped leather jacket" still turned up a result, and that result is relevant to what I'm looking for.

Anyways, all these searches turn up RELEVANT products to what a use would type in - working as intended.

Now, if you're a older-generation merchant looking to game the system, and now can't be found, once again I suggest that the merchants concerned should go back to their stores and revamp their listings to accomodate this.

"Green Striped Leather Jacket with Gold Tassels and Silver Buttons" would be highly preferred item to search than a bloated and frequently abused keyword lists that describes the same item as;

"green, leather, jacket,gold, silver, platinum,blue, banana, discount,sale,[brand name],jacket,coat,oranges, fishing,medieval,pockets,stitching,tassel,pom-pom,uniform,officers,etc"

Merchants need to upgrade their thinking, product listings, products, keywords and product listings away from 90's-style spammy SEO crap tactics, into content-based, descriptive and accurate titles and keywords. Most of the posters here complaining so far have numerous examples on their product pages that are pretty much straight up abuse and spam of the MP listing guidelines - they should correct their practices first and then see what the results they get.

Then we can continue the discussion whether the MP search is broken or not, because for me, it is not.

Related Resources:

The key to effective keywording is to

use all the terms

– 

and only the terms

–

that are clearly conveyed by the content and context

of the image.

- iStockPhoto

Think about the words users would type

to find your pages, and make sure that your site actually includes those words within it

. 

- Google Webmaster Guidelines

The written words in your shop can make it easier for shoppers and search engines to find your items

.

- Etsy Sellers Handbook

Appeal to humans, not search engines.

Be descriptive. Including relevant common phrases

in your listing will increase its visibility

.

- SL Guidelines For Creating Search Listings

 

[Caren Jewell]

@entity0x

what you wrote:

If a merchant doesn't like the results, they can always pack up their shop and leave...

or ..

Then we can continue the discussion whether the MP search is broken or not, because for me, it is not.

 

************

Good that all work fine for you and yes is better to continue it here - I thought the same after I read your respond

I agree that we as Merchants should go with the future - I be always open for new Ideas and ways - if its simple and make sense.  I wish all Merchants with problems here luck to find the balance for their business in MP - think we need it

(discussion closed for me too)

 

[Trip Hastings]

I'd like to throw in a commentary relevant to my own products and interest. I create for (and use) items made to work with the SMB mesh body. As such, I routinely search for the term SMB on Marketplace, sorted by Newest. Since the new Marketplace "upgrade", that search has become entirely useless. I have since learned that I can use the term "SMB" (note the quote marks) to get the results I want, but not everyone is going to figure that out. To illustrate exactly why this is such a problem, I'll present the following statistics.

- A search for "SMB" returns approximately 8500 results, and has done so reliably, increasing steadily only when new items are added. This is consistent with the pre-upgrade numbers, and returns 99% relevant results.

- A search for SMB done approximately 12 hours ago returned approximately 11500 results, with a very significant number of them completely irrelevant (a random sampling reveals absolutely no relation to the search term anywhere in the  title, description, keywords, etc.).

- A search for SMB performed about 5 minutes ago returned a staggering 20500 results, with the problems of the previous search obviously massively compounded. I had previously observed the number fluctuating up and down to a much smaller degree (no more than 1000, though that's still significant). This is simply absurd.

And, just to add one more voice to what others have said, my sales have most definitely almost completely dropped off since the new code went in. I don't keep sales stats, so that's purely anecdotal, but there it is.

[Codex Alpha]

Yet my search for "SMB" ( brand name or initials for something ?) turns up immediately 8252 relevant search results.

Appears to be child avatar panties, swim suits and pyjamas mainly?

Even then, assuming you have troubles, just entering "SMB" should have caused problems since it's one keyword, and initials at that - yet I still got relevant results.

Also, improving your search skills and including more keywords in your search should help you greatly

Examples;.

"SMB overalls" ( 415 Results ) would return more relevant results than only "SMB", and even more so if you searched for "SMB Pink Overalls" (13 Results ) or "SMB Pink Camo Overalls" (1 Result)

Working as expected, and intended.

Want more relevancy in your searches? Use more than one keyword. Want to be found? Use relevant and accurate titles that describe your product.

---

"Relevance : How closely the elements of your ad campaign match what a person seems to be looking for."

-

Relevance: Definition (from Google)

---

[ChinRey]

Oh well, one more post then. I hate to repeat myself but:

---

entity0x wrote:

Also, improving your search skills and including more keywords in your search should help you greatly

---

Your absolutely right. Only, how exactly are you planning to teach 900,000 SL users how to improve their search skills? It's a rather daunting tasks you have undertaken there. I can only wish you luck.

 

---

entity0x wrote:

Use relevant and accurate titles that describe your product.

---

Most people actually find complete sentences to be easier to read and more descriptive than a string of single words. Search single keywords higher rank.

 

---

Edit - three more factors:

Let's say I want a blouse with appliers for Maitreya. So I search for:

• Blouse Maitreya applier

and I get a list of titles saying:

1. Blouse Maitreya applier
2. Blouse Maitreya applier
3. Blouse Maitreya applier
4. Blouse Maitreya applier
5. Blouse Maitreya applier
6. Blouse Maitreya applier
7. Blouse Maitreya applier
8. Blouse Maitreya applier
9. Blouse Maitreya applier
10. Blouse Maitreya applier
11. Blouse Maitreya applier
12. Blouse Maitreya applier

Yes, I know that already. Presumably they wouldn't have been listed at all if they weren't blouses with Maitreya appliers. I want to know what's the difference between them but there's no room for that when the title field is transformed into the de facto keyword field and keyword dilution is so severly punished as it appears to be.

 

---

As for the keyword spammers, the only difference this shift from keyword field to title field will make is that they'll start entering all the spam into the title field. That isn't going to help much, is it?

 

---

And finally it's the danger of making too specific searches

I need a Victorian style mesh door for my house. So I search for

• Victorian mesh door

and get no hits. Because all the makers of Victorian mesh doors have learned to avoid keyword dilution so they're only listed as:

• Door

To see that problem from the seller's point of view. Let's say you have made a door that is very well suitable both for a Victorian and a steampunk house and you decide to aim specifically for those markets and ignore all those potential customers who skipped your search skill classes and only search for the generic "door". Then, if you want a decent ranking, you have to choose, do you want to list it as "Victorian door" or "Steampunk door" because you can't list it as "Victorian Steampunk door" or - even worse - "Vicorian or Steampunk door" or - worst of all - "Door for Victorian or Steampunk houses".

[Trip Hastings]

I feel like you have drastically (and perhaps deliberately) missed the point of my posting. This has nothing at all to do with a failure in my search skills, or anyone else's.

First off, I did say right off the bat that "SMB" gets the desired results (in this case, any and all SMB-related products).

The problem comes when I search for SMB, without the quotes. I expect a large number of result (in this case, the 8,252 you mentioned). I expect keyword-relevant results. Some variation between the two makes sense, except that the vast majority of results returned (at current search, 20,564) have ZERO apparent relevance to the search term. This conclusion is based on a random sampling of items chosen from the first 96 results of the search, sorted by Newest, selected from items that do not have SMB in the titles. The page source code was examined to get a full context of all descriptive elements, including keywords.

Understand, I am not LOOKING for highly refined search results for specific products in this search. Simply all relevant ones. I sort by Newest to find whatever has been most recently released. Under default settings, I must sort through several pages of results before I find the very first one that in any way relates to my term. In the previous iteration, enclosing the term in quote marks was not required. I gather that when searching without them, the back end uses a fuzzy logic routine to determine possible alternate related terms. It is clearly prone to being overly broad in scope when a single term is used.

As a test, I tried a few other single word search terms. The term computer, for instance, returns 20,502 results, while "computer" returns 3,536. The term house returns 320,058 results, while "house" returns only 145,513 results. The term soda returns 5,451 results, while "soda" only returns 2,050 results. In each case, the top results when sorted for Relevance (yes, the default) were similar, but other sorting methods gave very different output.

[ChinRey]

---

Trip Hastings wrote:

The problem comes when I search for

SMB

, without the quotes. I

expect

a large number of result (in this case, the 8,252 you mentioned). I expect keyword-relevant results. Some variation between the two makes sense, except that the vast majority of results returned (at current search, 20,564) have ZERO apparent relevance to the search term.

---

That's probably because searches without quotes use fuzzy matching and with a keyword like that things can get really fuzzy. Just take a look at the suggestions the spellchecker on this forum has for SMB.

[Freya Mokusei]

---

Trip Hastings wrote:

I don't keep sales stats, so that's purely anecdotal, but there it is.

---

FYI - it would still be anecdotal even if you had kept sales statistics. ;-)

Anecdotal means that it how a situation applies to you may not be the way felt across the board. Your sales have dipped, but this doesn't implicitly mean that everyone's sales have dipped - therefore dipping sales is anecdotal.

I'd agree with Entity in this case, expecting a three-letter acronym to be a useful keyword wouldn't ever be realistic. Your customers are unlikely to search in this way (rather than all your stuff appearing, they're more likely to view through your store page), you're far better trying to improve relevance to reach new shoppers.

Not criticising you, and I've not used Marketplace in probably a year - so I've no perspective on the new search stuff. I mostly like statistics, but SEO is something I have a knack for.

[ChinRey]

If I can change the subject slightly. Here's what the autofill function wants you to search for if you're looking for a door:



That brings another factor into the picture. Maybe it's better to just ignore search ranking completely and instead try to figure out how to get your titles included in the autofill list?

 

Edit (agin): Turns out it's very much biased towards old listings. So rather than try to get your own unique titles listed, just copy the ones that are already there.

[Trip Hastings]

Yes. Anecdotal as it relates to me, based on my general recollection. As opposed to concrete statistics of my personal sales. I wasn't claiming to speak for everyone. Geez.

Edited to add:

Except it's not (just) a three letter acronym. It's a descriptive term for a (relatively, within a niche group) major product with many users. Perhaps you could take the time to review my second post, in which I posted statistics for other single word search terms.

[Trip Hastings]

---

ChinRey wrote:

That's probably because searches without quotes use fuzzy matching and with a keyword like that things can get really fuzzy. Just take a look at the suggestions the spellchecker on this forum has for

SMB

.

---

My suggestion being that perhaps the fuzzy matching is OVERLY so when singular terms are used.

[Freya Mokusei]

---

Trip Hastings wrote:

As opposed to concrete statistics of my personal sales.

---

This would still be anecdotal! Any claim that is based from your experience (all of your claims in this thread - I wasn't saying otherwise) would be anecdotal.

Fuzzy matching on short search strings is intentional. It's considered poor behaviour to not attempt to complete a "short" query. It would be rare or unlikely that a three-character search string would be seen as 'complete' - precision is required.

"SMB blue shoes" is objectively a better search string than "SMB" - it tells the engine how to best satisfy the user and allows creation of a diverse results page. This is better because it enables the user to drill-down, rather than stifling too early.

SMB might be a useful term to you as a human. It is not a useful term to the search algo, it's too short and non-descriptive. Yes, explicit searches (those with " ") will return more specific results, but these are also more stifling (e.g. maybe you're actually looking for a laptop type computer, not a "computer", even if you search for computer), which is why this is not default behaviour. Search algos typically assume the user is searching inaccurately, not explicitly.

Hope this helps.

[Trip Hastings]

---

Freya Mokusei wrote:

---

Trip Hastings wrote:

As opposed to concrete statistics of my personal sales.

---

This would

still

be anecdotal!

Fuzzy matching on short search strings is intentional. It's considered poor behaviour to

not

attempt to complete a "short" query. It would be rare or unlikely that a three-character search string would be seen as 'complete' - precision is required.

---

The statement was anecdotal vis a vis the overall "marketplace sales are down" narrative, regardless of statistics. But in reference to an analysis of my personal sales, and that alone (which is the context I used to word in), giving statistics would have been hard evidence, not anecdotal. Why is this pedantry even being addressed? *headache* I won't mention it again.

On the actual topic at hand, if the search cannot handle a short term that appears in thousands of products' descriptive text, then I would have to assume it is not indexed properly. Your example of a spellchecker is telling. It only gives suggestions for alternates because it does not recognize the term.

As an aside, searching for smb (sans quotes) now returns 27,002 items, an increase of over 6,000 since I first posted a few hours ago. For whatever that's worth or means.

[Freya Mokusei]

---

Trip Hastings wrote:

Why is this pedantry even being addressed? *headache* I won't mention it again.

---

 Sorry, just because I'm a statistics nerd. It wouldn't have mattered if you'd presented hard evidence - if it only related to your sales it would still be anecdotal evidence because it's a non-representative sample. I'm not trying to be a pain, I'll happily abandon this point - it's just a word that caught my eye.

---

Trip Hastings wrote:

On the actual topic at hand, if the search cannot handle a short term that appears in thousands of products' descriptive text, then I would have to assume it is not indexed properly.

---

You'd be wrong. I've tried to explain this myself, but perhaps I'm not doing a great job. I'm far better at understanding this stuff than explaining it to others. Maybe this will help: Search Engine Watch - Longer Queries are Becoming the Norm and this one: Forensic Focus - Beyond Keywords

---

"Using short keywords or acronyms – including company names, slang words or a person’s initials – will often return huge volumes of irrelevant data. Such a short term will occur with great frequency within the data. For example, the letters “ge” occur more than 30 times just in this article."

---

It's not about frequency, it's about diversity of the search result - an acronym probably won't be correctly interpretted by automated means and so it jumps into fuzzy matching very very quickly. The aim of a Search Engine is to firstly help the user clarify their search - diversity enables this.

A more typical search route might go:-

• SMB (too wide, engine has no idea)
• SMB shoes (still too wide, but at least the engine can offer types of shoes)
• SMB shoes blue (narrower, now the engine can offer types of blue AND types of shoes - diversity is happening!)
• SMB heels dark blue zips (precise! The user now knows what type of blue AND what type of shoes. Probably leading to the specific item!)

It sounds like you're seeing this a lot, especially because of the variance in the number of results. What absolutely isn't happening, is a change to the search mechanisms over this short of a timespan. You're not seeing new changes each time you get a different number. You're just seeing a series of different bad guesses at what you probably mean.

Again, sorry for any confusion caused. SEO is hard, and while I'm good at it it's not something I'm fluent at explaining.

ETA: Some backup sources and examples.

[Trip Hastings]

---

Freya Mokusei wrote:

---

It sounds like you're seeing this a lot, especially because of the variance in the number of results. What absolutely isn't happening, is a change to the search mechanisms over this short of a timespan. You're not seeing new changes each time you get a different number. You're just seeing a series of different bad guesses at what you probably mean.

---

Perhaps I should have also asked YOU to not mention it again as well... anyway.

Actually, I believe what triggers the change in the number of results is new items added to the Marketplace that have the actual key term (or possibly also some close guesses) included. I've seen the number stay stable for extended periods (I look it up frequently when I get bored), and then shift. Previously it was a relatively stable range right above 10,000, only shifting in relatively small amounts when it did change. I only felt the need to point it out as a flaw in the system when it suddenly skyrocketed. For instance, in just the last short span of time, several SMB items were added, and the number has dropped significantly (though still very high).

I do actually understand the concept of a fuzzy search. The results being returned may well make sense to a machine, but as I believe you said, not to a human. That doesn't make the system less flawed. Ideally, the system should return results based on the way it is likely to be used, rather than expecting users to conform to an imposed syntax, as much as possible. Searching for a simple word should reasonably return results based strongly on that word, though that is of course my opinion.

If the system can't be relied upon to intelligently determine the closest matches and present them to the user, then it needs to at least provide instruction on how to improve the results received along with those results, on screen, at the time the search is performed. I do applaud the fairly clear and concise instruction provided on the "Help" link on the top menu bar on the Marketplace webpage, but from a user experience perspective it is perhaps somewhat insufficient in terms of presentation. In fact, as a Second Life user, I initially completely ignored it, as I assumed it would present the same page the Help link that virtually any other Second Life website gives (including this one), that being the general Second Life support portal. These forums, for instance, have a specialized "Community Help" link that is clearly unique and contextual to the use of the Forum.

Edited to add a summary:

You're seem to be clearly looking at this purely from a technical database and algorithm perspective. I am much more focused on the user experience angle. The simple keyword approach may not have been the most robust, but for very simple searches, it was much more intuitive.

[Freya Mokusei]

---

Trip Hastings wrote:

Searching for a simple word should reasonably return results based strongly on that word, though that is of course my opinion.

---

That wouldn't be considered a useful way for the engine to respond. It wouldn't let the user find similar words that they might instead be searching for (again, my example of laptop vs. computer). The idea would be that a search for computer could return laptops, and this diversity allows better targetting in secondary queries. I'm not saying you're wrong, I'm just explaining how search engines were designed - they anticipate that the user is just as faulty as the indexing mechanism.

They're not humans. Machine systems become flawed every time they encounter humans (at least until we do a little better at AI). Humans are adjusting to the way search strategy works (a vital skill!) but it's slow, and the SL userbase is not tech-savvy by average. I agree the service provider could make assisting the user more of a priority, but remember their background: In Silicon Valley, everyone knows how to Google - everyone who works on Marketplace probably understands syntax-based searching. This is probably a blindspot to them (blindspots are bad) - fortunately exposure helps, those who've used search tools for a while are improving.

I agree we're looking at different angles, I hope you've not read my posts as taking 'a side', there's opportunities for improvement everywhere. I'm explaining my perspective based on my background, which is... largely speaking to machines, and not humans. Keyword searches are an element that humans are moving beyond as well, and our marketplace ecosystem is too complex to support them (and has always been - 'SMB' is just a very good example of why).

[Trip Hastings]

---

Freya Mokusei wrote:

---

Trip Hastings wrote:

Searching for a simple word should reasonably return results based strongly on that word, though that is of course my opinion.

---

That wouldn't be considered a useful way for the engine to respond. It wouldn't let the user find similar words that they might instead be searching for (again, my example of

laptop

vs.

computer

). The idea would be that a search for computer

could

return laptops, and this diversity allows better targetting in secondary queries. I'm not saying you're wrong, I'm just explaining how search engines were designed - they anticipate that the user is just as faulty as the indexing mechanism.

They're not humans. Machine systems become flawed every time they encounter humans (at least until we do a little better at AI). Humans are adjusting to the way search strategy works (a vital skill!) but it's slow, and the SL userbase is not tech-savvy by average. I agree the service provider could make assisting the user more of a priority, but remember

their

background: In Silicon Valley, everyone knows how to Google - everyone who works on Marketplace probably understands syntax-based searching. This is probably a blindspot to them (blindspots are bad) - fortunately exposure helps, those who've used search tools for a while are improving.

I agree we're looking at different angles, I hope you've not read my posts as taking 'a side', there's opportunities for improvement everywhere. I'm explaining my perspective based on my background, which is... largely speaking to machines, and not humans.

Keyword searches are an element that humans are moving beyond as well, and our marketplace ecosystem is too complex to support them (and has always been - 'SMB' is just a very good example of why).

---

A search engine only knows "laptop" is a close equivalent to "computer" if it has been taught that it is so, whether by human hands or data analysis. If it doesn't know or can't know that X is approximately equal to Y, why would it ever need to present Y when looking for X? I am suggesting that the system's ability to make "bad guesses" is overly broad in some way, and may require refining. To refer back to an earlier point, I'd say a thesaurus might be a little more useful as a relevancy check than a spell checker. While you may disagree, that is my firm opinion.

I do, however, agree that user experience is very frequently disregarded by technology developers. Because as you said, machines are not human. Therefore it's up to humans to ensure they're properly designed to be useful TO humans. It's my sincere hope that someone with an eye for user experience sees all this and determines a way to address it. It may not be the most common use case, but it's a legitimate one. In a case like this, the Relevance sort option becomes the only one with any strong usefulness (though people seem to have strong opinions on how well THAT works now).

Perhaps if there was a search mode that weighted relevance alongside other factors (sort by age, price, etc) and presented the best results. That would be nice. Less accurate than simple keyword searching, but with the advantages of fuzzy search intact.

[Freya Mokusei]

Kohonen-model Neural Networking allows for engines to teach themselves the correlation between subjects - and is typically the technology being used (or, well, as a base). I use these regularly, they do have limitations - they don't correlate subjects based on purpose (or other human-recognised attributes), but instead on correlation between acceptance tests.

In short (and as a very indirect example), if a search engine offers you a product page for "carrot" when you search for "shoe", and you click the carrot... carrots are more likely to be associated with shoes in future. That's how it builds the association between "laptop" and "computer" - essentially through people getting their primary searches slightly wrong. This works to avoid the drift of words (kind of like a thesaurus!), but it makes it especially hard when a word has multiple connotations (e.g. acronyms) or doesn't produce 'like-for-like' results that can be used for testing (e.g. company names rather than descriptive names). "SMB" hits the sweetspot - it's both puzzles together.

You're right that - too often - people throw a search algo at the problem and assume that will sort out the mess on its own. If you look around the web today at some clothes shops (1, 2) you'll notice that they have far more available options when searching than just text entry - filter by size, by style, by colour, by gender. A more natural solution to navigating large datasets is construction of a tightly-designed search system that compliments the market that the site operates across and gives users the options they expect to see. Linden Lab did this once, for the original in-world viewer search system (now it uses a re-jigged Google ANN engine). Everything they've produced since - being derived rather than designed - has been less suited for Second Life.

Marketplace is an awkward beast, it's hard to search the same dataset for lawn chairs and rocket ships and dark blue heels and it's very difficult to build-in suitable filters. The history of the Marketplace service (being born out of acquisition / competition) also never allowed it to have development time spent on avoiding the mess of single syntax search. My suspicion is that getting them to write a new search engine this late in development is... not likely. I entirely understand the user experience argument - it's way more important than making developers happy. I've always wished Marketplace could be a better attempt - even IMVU's Catalog did it better (through virtue of it having been designed in since the beginning).

Syntax search maybe the best option we have - fortunately there's a whole industry designed around succeeding in this environment.

This has been a good chat. Thanks for understanding my perspective.