How stupid is this?

[Pamela Galli]

I have not had the stomach to look at inworld search for 6 months, but finally did. It was not pretty. Aside from the general mindlessness of the results I saw, I did not realize that now the results for the singular and plural forms of a word will be radically different!

 

Can you believe a search engine that doesn't know that a person searching for a prefab and one searching for prefabs are searching for the same thing???

 

I made a JIRA -- https://jira.secondlife.com/browse/WEB-4130

 

PS I also made another JIRA because when you click on the link in search to one of my prefab sims, nothing happens!

 

[WADE1 Jya]

Voted! & thanks for the heads up, checking my links now to ensure everything is working.

[TatianaDokuchic Varriale]

Thanks, I voted.

I noticed this a while back and adjusted accordingly.  I can remember (almost like it was yesterday though it must have been years ago) when I went through everything and removed the duplicates I was using to make sure I could be found using both "antique" and "antiques".  I didn't mind making the change then because it was a better system.  Finding out that this had been reversed really sucks.

 

[Darrius Gothly]

This is a side-effect of switching from a mature search system (like the GSA) to a D-I-Y Search Engine (like Lucene). The GSA included word-stemming in its keyword indexing algorithms. Lucene indexes everything separate and depends on proper construction of the query parser to handle such things as word stemming. It's a finesse item that hopefully they will add as they gain experience with the tool. (Assuming their "inevitable firing" doesn't come too soon.)

[nikita Jefferson]

If it was'nt for the fact i have my shop on my homestead which i have had for 2 years ( the homestead) i would give up my shop.

It seems every "improvement" to inworld search LL makes it just gets screwed up worse each time,i at least got some shop visitors before but since this last change,i barely get 5 a day.

The shop is more a hobby for me than a money making thing,i gave up on that silly notion a while ago.

I did file a ticket before the last search change because i was dropping in and out of search all the time,all i got back as an answer was to put a space after the comma in the description,so forget looking to LL for help and why should they ,the classifieds is what they want you to use,no money in search so they don't give a damn about it or us,they prob keep screwing up search to drive  merchants to the marketplace or to take out a classified ad,after all they can " tax " marketplace sales ,

 

[Ciaran Laval]

---

Darrius Gothly wrote:

This is a side-effect of switching from a mature search system (like the GSA) to a D-I-Y Search Engine (like Lucene). The GSA included word-stemming in its keyword indexing algorithms.

---

You make this sound as if GSA was efficient, it wasn't, and you make it sound as if keyword stemming worked well, it didn't, indeed we ended up with the amount of characters we could use in classifieds being reduced because of keyword stemming.

[Darrius Gothly]

---

Ciaran Laval wrote:

---

Darrius Gothly wrote:

This is a side-effect of switching from a mature search system (like the GSA) to a D-I-Y Search Engine (like Lucene). The GSA included word-stemming in its keyword indexing algorithms.

---

You make this sound as if GSA was efficient, it wasn't, and you make it sound as if keyword stemming worked well, it didn't, indeed we ended up with the amount of characters we could use in classifieds being reduced because of keyword stemming.

---

Sorry if I gave the wrong impression Ciaran. The GSA was ill-suited to its task of In-World Search. Overall the Lucene Engine stands a better chance. But it's not nearly the "Plug-n-Play" operation that the GSA was (although we all remember how painful it was integrating the GSA LOL).

Lucene is much more a set of Lego Blocks which must be assembled and customized in order to suit the exact needs of the Search Application. The Search Dev Team is deep in the throes of trying to understand all the parts and how they can be connected and carved to make a better search overall. (Of course "better" in their definition does not exactly match what we as Merchants and Residents believe .. but that's another issue.)

Word stemming, in and of itself, was not the cause of reduced Classified Ad space. Whatever reason was given (and I frankly don't remember the exact official cause), word stemming is simply the act of recognizing words with identical "stems" as the same word. Thus "prefab" and "prefabs" look the same to the keyword indexer and search retrieval processes. Where it gets complicated is when you have words with identical stems but radically different meanings.

My favorite example (and one that Sea Linden took care of early on) are the two words "Teleport" and "Teleporter". The former is a verb while the latter is a noun. They are not the same word even though they both have the same root stem. What happens is the algorithm that spots stems must also consult a dictionary of "not the same" words that negates the stem recognition and declares the words separate and distinct.

The GSA came with this mechanism built-in and with a starter dictionary of common exceptions. Lucene has no such native feature nor does it come preloaded with exceptions; it is up to the developer to build the stemming and stem exception logic and build the required dictionary. As of yet, I don't believe the Search Dev Team has climbed that high on the Lucene mountain to begin caring or addressing a feature such as this.

[Ciaran Laval]

---

Darrius Gothly wrote:

Word stemming, in and of itself, was not the cause of reduced Classified Ad space. Whatever reason was given (and I frankly don't remember the exact official cause), word stemming is simply the act of recognizing words with identical "stems" as the same word. Thus "prefab" and "prefabs" look the same to the keyword indexer and search retrieval processes. Where it gets complicated is when you have words with identical stems but radically different meanings.

---

Sea did do a lot of unstemming but she also pointed out at a user group meeting that stemming had led to classifieds holding the top spots for loads of different searches and you could end up with three words per word due to how it was implemented, it wasn't just stemming in terms of the root word, they made different words mean the same thing, which meant people weren't getting found when they tried to use different words, so for example houses and prefabs might have been considered the same thing, this buried plenty of classifieds.

[Darrius Gothly]

---

Ciaran Laval wrote:

Sea did do a lot of unstemming but she also pointed out at a user group meeting that stemming had led to classifieds holding the top spots for loads of different searches and you could end up with three words per word due to how it was implemented, it wasn't just stemming in terms of the root word, they made different words mean the same thing, which meant people weren't getting found when they tried to use different words, so for example houses and prefabs might have been considered the same thing, this buried plenty of classifieds.

---

Ahhh yes. Thank you for jogging my memory. Yes, word stemming caused a massive fuuuu ... ummm .. problem with Classifieds. But that should have been addressed by changing how Classifieds were indexed and searched. The solution of shortening the available space was one of those "left-handed" soltuions that may have sorta kinda worked, but didn't really fix the problem.

(And let me add ... I am left-handed, so I'm perfectly allowed to use that term. Please refrain from email bombing me about being anti-lefty. TYVM)

[Ciaran Laval]

---

Darrius Gothly wrote:

 

Ahhh yes. Thank you for jogging my memory. Yes, word stemming caused a massive fuuuu ... ummm .. problem with Classifieds. But that should have been addressed by changing how Classifieds were indexed and searched. The solution of shortening the available space was one of those "left-handed" soltuions that may have sorta kinda worked, but didn't really fix the problem.

(And let me add ... I am left-handed, so I'm perfectly allowed to use that term. Please refrain from email bombing me about being anti-lefty. TYVM)

---

Indeed, I'd rather they worked it like the marketplace whereby you had a keywords part and then a non indexed for search purposes description where people could be more creative with their adverts.

[Darrius Gothly]

---

Ciaran Laval wrote:

---

Darrius Gothly wrote:

 

Ahhh yes. Thank you for jogging my memory. Yes, word stemming caused a massive fuuuu ... ummm .. problem with Classifieds. But that should have been addressed by changing how Classifieds were indexed and searched. The solution of shortening the available space was one of those "left-handed" soltuions that may have sorta kinda worked, but didn't really fix the problem.

(And let me add ... I am left-handed, so I'm perfectly allowed to use that term. Please refrain from email bombing me about being anti-lefty. TYVM)

---

Indeed, I'd rather they worked it like the marketplace whereby you had a keywords part and then a non indexed for search purposes description where people could be more creative with their adverts.

---

I wish they'd just let us mark a Marketplace Product for inclusion in the Classifieds shown along with In-World Search.

[Pamela Galli]

The knowledge base on Search clearly states that stemming is operative -- that "dance" and "dances" will return the same result. It does not. It is a bug and a major one.

If you have weighed in on the JIRA, please do.

[Josh Susanto]

Is it a bug or is intentional misinformation?

If you had, as the Lindens do, information on which word gets more queries, you could use that to personal advantage.

One such instance might not be enough to make you care, but how many other pairs of important similar words are there that will be searched for at different rates?

If you had the whole list, as Lindens do...