Traffic Alternatives - L$2000 Reward

Very similar to the proposal to only count premium accounts. It could potentially work if the identity verification was actually valid. Others have shown that it's easy to get bogus information to be accepted, and at times it can be very hard to get actual information to verify properly. If you only counted 10% of PIOF avatars, each one that gets counted would count 10 times as much as an avatar now. You could guarantee an average of 1 counted avatar 24 hours a day by faking up a bunch, logging in 10 at a time, and changing them out every couple of hours.

Just because a place is commercial doesn't mean that it is selling products. You can sell entertainment. You can have pay-to-play games. Whatever method is used to sort search results has to work for those places too.

The traffic count is still useful for people who don't game it. It's like a poor man's visitor counter. Removing/reducing its affect on search rankings is fine with me but I don't want to see it go away altogether. I'd rather it be enhanced to show more useful information.

--Hugsy

I was thinking that only 1% of the logins would have their traffic data accounted for. Traffic numbers would drop dramatically (be roughly divided by 100?) so use the past 30 days worth and call it "Traffic for the past month."

--Hugsy

But traffic is a fine and usefull metric EXCEPT for the businesses who are gaming it. My point is - none of those other places are gaming traffic - so the traffic sytem as is works just fine. There is only one reason to game traffic and that reason is to increase product sales. Remove that reason and there is no longer a reason to game traffic and the system works without need for changes.

1% of premium accounts is too small to get a good sample, even if you spread it out for a month. That's the percentage I had in mind when I said that if we could run a million concurrent and over 100 in a sim you would have the winning idea. Even 1% of the total population would be too small at this point. The smaller the sample, the higher the motivation to add a bunch of bots. A place that had 100 real visitors that each spent an hour a day would have huge traffic under the current system. In your proposed system, they would average 60 per day. Adding 24 groups of 10 bots that switched every hour would average 144 even if nobody else showed up, or 204 if the current customers kept coming. Someone with 240 bots could camp their own store and sell the same service to 9 other stores. Looking at how much people will pay for classified ads, it could be a very lucrative business model.

Entertainment venues can game traffic to be found the same way that stores do. Gaming venues can game traffic the same way that stores do. Just because you aren't selling a product, doesn't mean that you aren't trying to make a profit. And many of those places DO game traffic.

The reason I think the current traffic system is acceptable is that it's easy to avoid the places that are obviously gaming to a ridiculous level, and it's equally easy to avoid the places that are obvious ghost towns. All just by glancing at the list of results.

OK, now here's the point:

You don't need a general search ranking system. You need a mechanism to find games that you like, and a mechanism to find beaches that you like.

So the next questions are: What are the properties of games that you like? And what are the properties of beaches that you like?

No matter what properties you include in your search, you still need a metric to order them by. The current traffic metric instantly shows you which are extensively gaming their traffic numbers, which are always deserted, and which have a good chance of having the level of popularity you are looking for. It's easy to type in the name of the game you want, but when you sort by traffic, you can easily avoid the places that will always be too crowded and laggy, and also avoid the places that nobody spends time at because the games aren't worth playing.

If you're not going to attempt to actually answer the original question, please run along kid.

A recent blurb on my ACM new service said that some types of CAPTCHA are easily broken with some intelligent OCR. Some types are just downright dumb (e.g. one that selects from a store of a thousand images - perhaps a day or two of cheap labor to build a database that solves it.

I am trying to answer the original question. But to do so takes more information.

There's a well-known model for choosing metrics called GQM - Goal - Question - Metric. It's really simple: Before you can actually come up with a metric, first you need to come up with a good, well-defined goal, and then you need to come up with specific questions that help achieve the goal. Only then are you ready to look at metrics.

So I'm trying to elicit your goals. That's harder than it sounds, and more than simply asking what your goals are. That's partly because most people aren't used to putting their goals into explicit words, and partly because there are a lot of bogus assumptions that have been made throughout the entire history of traffic metrics.

Let's get to the meat of the matter. Remember, the only metrics of interest are how closely it matches your goal.

While you've implied that you're interested in popularity, you haven't explicitly stated it. The way it's been mentioned makes me think that it's there out of habit or by default, not because you really wanted. I have yet to see anyone, not just you, bother to explain the relevance of popularity to their search, in a critical way. My suspicion is that most people think that popularity implies that other people have found something worthwhile - a conclusion that's simply not valid in any general sense.

In fact, it's quite obvious from your remarks here that you don't want it sorted by "popularity". You explicitly said you don't want places that are crowded, and you don't want places that are empty. And in the first case, there's at least a plausible connection between popularity and crowds, and we know there's a connection between crowds and lag.

But if your concern is lag, then the question to ask isn't how popular a place is, but how laggy it is. For simplicity, I'll say let's filter based on average Sim FPS. This needs to be more refined (e.g. average Sim FPS over the last N days, or maybe even average Sim FPS between 2pm and 4pm over the last N days excluding weekends). But you get the point. For client side lag, you may also want to know how many bytes of texture need to be downloaded, perhaps on some coarse scale, perhaps how many flexiprims are there, etc.

At the other end of the spectrum, i.e. unpopular sites, you're making a couple of leaps. There are lots of reasons a place could be unpopular. The most obvious isn't that the games aren't worth playing, but rather that nobody knows about it. The other flaw is the belief that just because other people may not like it, you won't. (If other people's opinions were all that mattered, why aren't you spending all your time playing Warcraft instead of searching SL?) It makes more sense to first try to categorize games by what you like - cards? skill? reaction time? strategy? And so on.

Ultimately, though, there may not be a metric that properly describes what sort of games you like. In that case, random is as good an order as any. For the sake of comparison, how many people choose which movie to see based on box office results? And how many based on reviews? Granted, there's a different scale factor involved (at most a few dozen movies to choose from, instead of thousands, at least for theatrical releases). But the principle is the same: There is no general metric. At best, there's a personal metric - such as your favorite reviewer's rankings.

my point was, why accept a L$2000 reward when i can do better with things as they are?

think about it. people are making $7US per day while doing absolutely nothing, which equals about $210US a month.... times however many bots someone has.... let's say 4. $210 x 4 = $840. how about 4 more? and 4 more? and 4 more?

maybe NOW you see my point?

I'm not convinced that 1% is too small of a measure. The importance is to make sure the sampling is truly random. Going back to the Nielsen ratings, I tried looking for the number of homes who have TVs and found this on the Wiki:

http://en.wikipedia.org/wiki/Nielsen_ratings

It says, "there are an estimated 112.8 million television households in the United States". I couldn't find that number on the Nielsen web site. Their site does say, "Each year we process approximately 2 million paper diaries from households across the country for the months of November, February, May and July — also known as the “sweeps” rating periods."

http://www.nielsenmedia.com/

Inside TV Ratings -> Meters & Diaries -> Our Measurement Techniques

They also use electronic monitoring but could find a number for that. Regardless, that's only 1.8% and seems to work fine for them.

Of coarse the traffic numbers will drop but that's ok because it's the same for everyone. People will have to adjust to what the new numbers mean because it's different.

The fact that lowering the percentage will cause an increase in bots is the killer here. That's why you need a strong CAPTCHA implementation. The idea isn't perfect, but I think it would be better than what we have now.

--Hugsy

If you go back to my original post, I lay out 3 criteria. A measure of the popularity is the first one.


From the post that you quoted this time:
"The current traffic metric instantly shows you which are extensively gaming their traffic numbers, which are always deserted, and which have a good chance of having the level of popularity you are looking for. It's easy to type in the name of the game you want, but when you sort by traffic, you can easily avoid the places that will always be too crowded and laggy, and also avoid the places that nobody spends time at because the games aren't worth playing."
My specific goal when looking for a gaming place is to find one popular enough to draw a reasonable crowd while avoiding the ones that grossly abuse campers and/or trafficbots. The easy way to do that is just like I described, and requires the list to be sorted by popularity. Just like I said.



When I'm looking for games, I'm trying to get paid. If people can play at a place and make money, they'll play there.

Continuing to ask about the specific qualities that I personally like for all the different places that I personally search for will lead to one of two logical conclusions. You'll end up explaining how to use more search terms in the search window as if I don't know how to use it, or you'll reach the conclusion that we'd need a different search window for every combination of type of place and search criteria that people want. Neither of these has anything to do with the original question of producing a better metric than the current traffic number that could be applied to the Places search.

1% of millions would be awesome. 1% of 100,000 spread across several thousand sims is going to be way too low. When, not if, we get SL's numbers way up there, this becomes a much better solution than it would be now.

It's actually very different from the premium account suggestion.

And you're confusing identity with age verification. If you put your credit card or PayPal account on file, you're confirming your RL name (or the cardholder's in any case). The only way to cheat there is to commit identity fraud and I'm assuming most store owners aren't that desperate for traffic to resort to actual, true crime.

You also missed the point that each identity only has one avatar that's eligable for traffic. Logging in 10 alts wouldn't change the fact that only one of them is counts, the 9 others don't count at all so alts/bots are out of the picture for the most part so you're left with manipulating humans.

Capping the maximum contribution to traffic for a specific parcel per avie every day will make camping less attractive for people who camp and less effective for the person who runs the camping site. If you limit it to 3 hours per parcel per day maximum you shouldn't hurt too many legitimate ventures while camper sites suddenly need 8 times more unique campers per day from a smaller pool of people.

Camping will still skew the results, but far less than it does now. In the end the only close-to-perfect solution would be for LL to want to condemn gaming search with the penalty of revoking search listing privileges after one warning, but we all know that's not going to happen so likely the best solution will only be marginally better than what we have now.

From what little I know about the rating systems already in place; the rankings are coming from a formula with few variables. Only one or two at the most to determine popularity. These variables typically include the number of people during a given time and the length of time spent by each person (or bot, camper, etc etc). After putting on some controls, such as dropping the ones that only stay for a few seconds and repeat visits, you have the magical number. Is this representative of the current model?? If it is then the fomula is too simple allowing people to exploit it and take advantages.

Seems that the formula needs to be more complex with more independent variables to make it harder to exploit. Seeing how everything in this virtual world can be be programmed, categorized and easily referenced I offer this:

Camping numbers could be easy to eliminate by including, for instance, whether or not the visitor interacted with the environment (keyboard inputs, mouse clicks, number of unique items clicked on) as most campers just sit and do nothing. This could work in both stores and clubs. Most people click on things for purchase, info, notecards, to sit down, to dance, to interact with others.

One could combine this interaction data to each unique visitor to come up with something that might not be perfect but could zap most of the bots. The key would be to filter out any scripted events. This data could also be weighted depending on the type of business involved. Merchandise areas tend to have more items to click on than clubs. So you could weight mouse events more heavily for stores and keyboard events for clubs.

There are probably ways to get bots to walk around an click on things but being someone that is rusty on scripting I don't know how to do it.

I seem to remember the number premium accounts per person IS LIMITED to 5 per person. Maybe that has changed when I wasn't looking.

As for traffic being a usefullmetric it really isn't. Hits on your search listing that result in actualy TP's would be far more useful to a business owner.

Much though I dislike camping lag and bots. One thing that needs to be considered is that Camping does serve a purpose beyond lining the pockets of bot runners.THere is a fairly high percentage of new players that use it to fund thier initial explorations of SL. This gives them a chance to decide if it is something they really want to invest in or not. Many eventually do put payment info on file andbecome paying customers. But they will tell you having that period where they could get a little cash to start with was crucial to thier decision to stay. We do need more real people behind the AVs and less bots. When LL did away with sign up bonus for any but premium they pretty much forced many new players to camp and fed the bot problem.

You still misunderstand the point. In the original post, you laid out criteria for some metric. You said you wanted a metric of popularity. But you didn't say you wanted to find popular sites or state anything about how popularity was important to you. And that's my main point: You've made assumptions about popularity to hardwire it into your metric, which is the wrong way to do it.


The first is only sometimes true, but it's inadequate. You don't want people who are gaming their traffic at all - because that makes the traffic meaningless for your purpose. Or to put it another way, if this became the norm, then the people who are gaming their traffic would just have to moderate their gaming so that you couldn't tell this way. The second may be true, but as I indicated before, it's not a valid concern. Being deserted means that people haven't found it, it doesn't mean you won't like it. Either way, you're doing a job the computer should be doing. If it's something that's so easy for you to filter, then the search system should have filtered it for you already.

The third is more interesting, because for once you've phrased it in the positive and not negative. But is it true? Are you really looking for popularity? Or are you using popularity as a surrogate for identifying your real criteria?


Is the crowd important to you? Do you like the game better when there are other people around? Or does it indicate something else?


So what you're saying is that there's no way to search for games that have payoffs, so you're using popularity as an indicator. Is that right?


None of the above.

The only conclusion I'm leading to is that popularity is only being used because LL made the mistake of putting it there in the first place. It will always be possible to game popularity, and therefore it will never be reliable. And there are other, better ways to solve the problems that you think it solves.

[quote
Neither of these has anything to do with the original question of producing a better metric than the current traffic number that could be applied to the Places search.[/QUOTE]
If you want a better metric, you have to begin by removing popularity and traffic from the requirements. Start with a clean slate. Just for the moment, set aside the emphasis you've been putting on eliminating places that are either gaming the system or are empty. Instead, impose the constraint that you can't use anything that can be faked without being caught and potentially punished. (This excludes lying in the description, because that can be caught.)

Personally, I think the tags suggestion is a good one. But frankly, there's a reason I use secon-man.com as my primary way to find clothes, that I us SLexchange to find other items, and that I use word-of-mouth or other review sites to find interesting places. Popularity never enters the picture.

The final part of the constraint isn't very realistic though unfortunately .

Anything that results in more work for LL (investigating claims of "metric faking", sanctioning, following up, etc) will likely get dismissed with "won't scale, so no".

If it's subject to faking (most everything is) it needs to be limited in effectiveness, or self-regulate/self-adjust *somehow*.

Perhaps I wasn't clear enough, but this is backwards. The constraint isn't that LL actually do anything. It's just that in defining a metric, you avoid anything that's easy to game anonymously or cheaply.

If the main complaint about traffic is that it encourages waste of sim resources (because bots are used to game traffic, and in turn the bots lag the sims), the focus on traffic is misguided.

The main problem about the various means that Linden Labs has implemented to discourage waste of resources (limiting accounts, limiting prims) is that they don't directly relate to the things that actually cause lag.

Right now, the main limit I have to the way I use my little 512m of land is that I can put a maximum of 117 prims on it. That really does not limit well the stress I can cause for the entire sim. 117 wooden boxes does not cause nearly the sim strain that one box with a high-res texture and a script can cause to a sim. Heck, without even rezzing a single prim, my prim clothing attachments probably cause more strain on a sim than 117 wooden boxes.

If Linden Labs were to find a way to limit total resource usage per parcel (taking into account not just prims, but scripts and textures), then Linden Labs would be on to enforcing some true grid sanity.

I'm not terribly deep in technical knowledge about a program like Second Life. However, I know with MUDs, for example, there were ways to keep track of the size of objects in databases, the stress that some script functions put on processing capability, and limit them per user or area. That would lead me to believe that it would have to be feasible with Second Life.

I think that a low percentage as a sample would actually be a wonderful thing.
1. It would kill the bot trade (yes! - see below)
2. It would introduce some much-needed and essential variety into the range of places that people were pointed at.


1. The bots
People have argued that the response of the bot-runners to a low percentage of random logins being measured would be to flood us with even more bots.
It doesn't work that way however, not in SL.
Take for example a place that run 10 bots in a box to game traffic. If there was only a 1 in 100 chance that any one of those bots would count towards traffic, the the owner would need to run 100 times the number of bots to achieve the same effect.
So that would be how many bots on the parcel???
Ooooh! ...put on some 0's .... divided by your age plus the number you first thought of.... for approximately ...... eh .... about ....
....
1,000 !!!!
..................Ha-looww-ooooowww

Oops!

Well, if they owned the entire sim, and it was on a PI, then they could set the server to allow up to 100 avatars concurrent.
If all those avatars were bots, then with a 1 in 100 probability, they would have the same effect as would 1 bot currently.
Still a teensy weensy problem though ..... no room for any real people!!!
So reduce the bot count down to .... 80? and have the same effect as 0.8 of a current bot.

So:
On the Map
- the yellow bits are for sale
- the purple bits are for auction
- the green bits are bot-runners getting a traffic value equivalent to about 0.5 to 0.8 of a single current bot.




2. Variety

There's something broken and Darwin-denying about the idea of being "the most popular" in SL.
A whole sim has issues trying to deliver a decent experience to 30 avatars.
Two or more "most popular" builds in the same sim would be a nightmare.
Many people, myself included, learnt to avoid the top listings in search when they are associated with very high traffic counts.
1. We know they are going to be laggy as hell
2. We assume that they will actually be full of campers/zombies.

A SL in-world brand would really need to establish a chain of locations spread over a number of sims to be able to support the traffic that would result from actually being 'the most popular'. Think "SL Welcome Area".

"When, not if, we get SL's numbers way up there,..." nobody will want their build to be permanently at the top of "most popular".
Imagine the brokenness if 0.001% of 1,000,000 people searched, found you and tried to TP in.

A larger SL will absolutely need to introduce randomness into search result sequencing.
However, as has been well put, a completely random sequencing not based on some traffic metric would generally produce a list of BIAB outlets.
If the 'traffic' metric used to sequence results were based on a small random sample of logins (over whatever time period) then we have randomness, but based on real activity.
Even with that, the results presented in response to the same simultaneous searches would have to be based on a random/cycled selection of the appropriate candidates in order to avoid the problem of "too many " people trying to get into a location.

Goggle also has something called the "Goggle Slap". if you try to game there system you get bounce out of search all together.

The system as it works now is, OK, but it just need one addition, the "linden Slap", game the search with bots and lose you listing.

I am sure others would catch on real quick after the first few got delisted.

This wouldn't take much time for a Linden or a few Lindens to check the top results and green dot hop, just to see if the traffic is legit. It really only effects the top 20 to 100 listings .

i can visit 20 Sims to check that in about 30 min to an hour. so its probably less trouble than trying to code a solution for a social problem.

Just use a "Googlebot Linden"

They wouldn't need more bots to achieve the same effect though, they'd need the exact same amount since everyone else's traffic shrinks 100-fold as well.

In reality the averages won't play out quite so neatly and you'll get fluctuations day-by-day, but on the average nothing changes since everyone's traffic shrinks by the same factor (more or less).