New Traffic Calculations

What do you think???

Changes to the Traffic Scoring
Posted by Jack Linden on Aug 21, 2009 3:14:36 PM

Every parcel of Land on every region inworld has a Traffic score. Traffic is a measure we developed some years ago to show Land owners how much time other residents spent on their parcels. It's been used in various ways throughout our History to help reward and highlight popular places. More recently it became a part of the way we rank Search results.

This was fine whilst we had fewer regions and lower concurrency but as we approach 30000 regions calculating the amount of time that every resident spends on every parcel they visit, every 24 hours, has become an increasingly heavy burden on our systems. Over recent weeks the Traffic calculations have sometimes failed to complete during the 24 hour cycle causing frustration for Land owners. So we have been redeveloping the way that Traffic is being processed. The aim has been to make it more reliable and scalable as we grow.

Along with improving the process to remove load from our core systems, we have also taken the opportunity to improve the calculation itself. Liana will be posting more details on the specific changes on the technology blog; the new Traffic score will be simpler, more granular and therefore more accurate than previously.

Traffic will be the cumulative minutes spent on the parcel by all visitors to the parcel within the previous day. It will be calculated from the total seconds spent on the parcel, divided by 60 and rounded to the nearest whole minute (up or down). It is calculated in real-time, and will be updated every morning at 1am PST for the previous day.

Over time we expect the Traffic scoring used for Search ranking to develop away from the measure used by Land owners; it makes sense that the two diverge and the Search team will be talking more about their plans on their blog.

We will be moving across to the new Traffic scoring process on Monday 31st August, which means that the Traffic scores you see on Tuesday 1st September will be the new ones.

For the large majority of parcels the impact of this change will be negligible. Your traffic score may be a little higher than usual but your ranking relative to other parcels will remain close to where it is now. A small number of parcels may see a drop in Traffic score, especially if those parcels are visited only briefly, for example where residents are just flying through on the way to somewhere else.

Lastly, we realise how important our continued efforts to deal with Search gaming are. The announcement has had a clear impact but there are still venues out there that are gaming the Traffic score so we will be pushing harder to make sure that unfair use of Bots or other means to game Search are stopped.

I think it's all smoke and mirrors. I can't see the first thing in the whole writeup on it, in either of the two blog entries, that is one iota different from what the traffic calculations have been for many months. it looks more like somebody who's not up to speed on changes that were made a very long time ago and thinks this is something new they need to tell the public about.

Meet the new boss, same as the old boss......

or they think we are all stupid, nothing to see as nothing has changed, unless of course they disconnect traffic from the search then that will be a decent change IMHO

And whats the news? What really changes? Its still magnetic for persons who want to manipulate the traffic

All the new changes, from adult, over search, over xstreet, over content seller programs, over traffic score ideas, over whatever and up to whatsoever - all this is running - from my view - under one single headline: customers and comsumers are incompetent in LL's eyes.

So LL's new theology and evangilism of biz is:

"In god we trust" (like to read on the backside of every dollar)

"In Brand we trust"

"In traffic we trust"

"In censorship we trust"

"In FIC we trust"

"In labels we trust"

"In Neuro-linguistic programming we trust"

"In hubris we trust"

And that's their new bible or "TAO".

In 2 things they don't trust: the competences of creators and customers.

For this information we pay up to 295 dollar a month, just to hear every day, that we are: too stupid for everything.

So, calculating traffic is a load on their core systems and they still insist on calculating it? If they made it something you had to opt in to, like appearing in places search, then they could avoid calculating it for most parcels. That, and removing it from calculations for search, would seem to be the highest-quality solution.

Instead of telling us how they'll be calculating it I wish they'd explain why they calculate it at all.

The change seems to be that if an avatar is only on a parcel for 20 seconds then they only score a third of a traffic point. I'm assuming people were teleporting in and out to up traffic scores which would put a load on the server.

As for counting traffic, I suggested some time ago that I should be able to opt my residential parcels out of traffic, there's no need for a traffic count on a residential parcel.

Traffic counts are worthless. Save the cycles and eliminate them.

What makes traffic so easily gamed is that it is the *only* numerical search ranking we can see. If we had another value, like number of unique visitors, that would give us more information to use when selecting places to visit. Traffic of 10,000 made up of 500 unique visitors for 20 minutes each vs 7 visitors for 1440 minutes each gives totally different impression of real popularity.

Ideally I would like to see several checkboxes: "word match", "traffic", "visitors", "parcel size", "parcel age", and then allow sorting by columns or by a logical AND function of which ones are turned on: Sort by Word match AND parcel size AND parcel age = I want to find a place that matches my search term, is large, and has been there a long time.

Something like that would make search more flexible and useful. If you only have one "score" to rank people, then they will work hard to come out on top in that measure. Having multiple ones, especially ones that cannot be changed easily (parcel size and age for example) makes it harder to game, and easier for us to search the way we want to.

Neither parcel size nor parcel age seem particularly useful. What is their value to you?

That is brilliant. I hope someone up at LL takes notice.

As dude said in the example, he was looking for a place that's stood the test of time. And as far as parcel size, that can also contribute to your ability to get an idea of a place before you spend your time TP'ing to it.

Size would be useful in a sandbox or rental search. Age would be useful in selecting a business that will not likely poof tomorrow. I listed those particular ones because the data already exists for every parcel anyway, so would not require added tracking.

My main point however is not *which* columns of data are available. It's that more than just the one of traffic score be available to sort with, widening the usefulness of search.

I like the idea.
I am sick of botters and their unfair traffic.
I do think good product and having a community group that you keep up with and really make it more then a simple update speaks all the numbers you need.

I find people that bots have a low product quality over all,when i look in search at the high traffic places and visit all i see is a mess of reseller products.

There is a difference in the new calculations. They will be doing it by seconds rather than minutes, and then dividing the total seconds by 60. It means that a few seconds walking across a parcel will contribute a few seconds rather than a point for a whole minute. The results will be marginally different but it is different. Perhaps doing it the new way will be easier on the system, but it won't make any difference to the benefit of using bots. A bot will still add 1440 points in 24 hours.

I'd like to opt my workshop out, because honestly, it's not in the search and I tend to be the only one there.

The big change is in the method of calculating it - which should have the end result of a major drop in lag if I understand it properly.

What I'm unsure on is the method that was used previously in the formula for traffic. However the manner - doing it all at end of week - is obviously problematic.

As noted, it often failed to complete - which meant that some people never got any traffic score for that week. Now, spread out in real time, it will manage to capture everyone's actual traffic.

This post isn't about how it will prevent gaming, but how it will become accurate. From there, they're going to need human 'Lindens' running around to catch the people using bots to game it...

As long as there is a traffic score, there is nothing that can automatically prevent a bot or camper from gaming it. You could enable some people to 'own up' and voluntarily opt out bot / camp avatars from traffic scoring - but that only reduces traffic of honest people.

Humans are needed to catch dishonest people - and with 30000+ regions, it would take an insane number of humans to catch all of the problems. Even if they merely focused on the top 5% of traffic scores... people gaming the system would just stop when they hit some point - "throttling" themselves up and down to stay under the radar but still high in search.
- and 5% of 30000 is still more than you can reasonably employ people to catch.

- A lot of people use various scripts that record unique visitors. Including that into the search engine itself is a golden idea. I really don't care who the people in the list I get are, just the number. That said I also like knowing which ones return, something my present script doesn't tell me.

Some possible other useful numerics:

1. Number of search hits this entry gets.
2. Number of hits its gets that result in a TP to the parcel.

These ones have the obvious flaw of missing hangout places:
3. Number of times a linden transaction happens on the parcel for a value more than 0L.
4. Above but including 0L transactions (freebie stores).
5. Above two minus any entries from the same IP-address.
- and there would need to be more methods in place to prevent results from a series of bots simply handing money back and forth in a circular network. Given a day of brainstorming, somebody could fix the flaws in this.

You could use the above for places flagged as shops. And for places flagged as hangouts (and the other social categories):
6. Number of times a unique message appears in local chat that is avatar created (not a gesture or script).
7. Number of times a calling card is exchanged or a friend request is accepted by persons who have never before done so with each other.


-Etc...

Some of those have potentially major flaws. Its just a 5-minute brainstorm.

Don't go all out attacking them and derailing things... Just take it for a brainstorm please.
- I know people like to lock onto and zerg-focus on one flawed idea to the exclusion of the whole central point of a post... so brainstorming is always dangerous on the internet... But that -IS- a brainstorm list, not my opinion of the Holy Gospel. So if there's something in there you think is flawed, just move on with a better item.

With some thought, the flawed ones could be fixed or weeded out, and I'm sure other people's brainstorms could generate more ideas, and with some work - the best 10 or 20 such ideas could become a new set of numerics.