Wooolf! I mean re-rank.

[Joseph Clemson]

BHZ is a bit like trainspotting. It has little relevance to reality but it enables people who have not enough to do to discuss its lack of relevance with other people who have not enough to do.

 

Alan

 

Ah Alan, I would beg to differ. The recent downgrading in my BHZ placing is proving to be a good indicator of my placing in 'real life' test searches, which I also check from time to time. In these I was routinely in the upper quartile of results, and now I find myself in the lower quartile.

 

And yes, I'm a trainspotter too!!   

[John Mitchell]

 

As mentioned, this is the first time I've tried the BHZ thing, and I'm wondering if there is an obvious correlation between the number of images in a pseudonym and BHZ ranking -- i.e. do pseudos with more images always show up higher in BHZ searches than those with fewer images?

 

It's usually the opposite: have more sales and zooms, but keep your collection small.

Remember that a healthy pseudo does get not that much zooms. It will get a lot of views though, which seems to get punished by the current algorithm.

 

wim

 

 

Not in my case, Wim. My pseudonym with about 4400 images appears on page one, while my secondary one with only 500 images shows up on page 21. I didn't really expect my secondary pseudo to have a high rank, but I'm surprised by how big the discrepancy is. I believe a couple of other people have mentioned this as well, so I'm wondering how accurate BHZ really is. Perhaps it ranks individual contributors' pseudos on the basis of size more than anything else. Because I have only two pseudos, I can't really determine if this is the case. Those with multiple handles might, though.

[John Mitchell]

BHZ is a bit like trainspotting. It has little relevance to reality but it enables people who have not enough to do to discuss its lack of relevance with other people who have not enough to do.

 

Alan

 

Ouch! You must be having a really bad day.

[Pearl]

BHZ is a bit like trainspotting. It has little relevance to reality but it enables people who have not enough to do to discuss its lack of relevance with other people who have not enough to do.

 

Alan

Sorry if you don't get it.

 

Pearl

[Sultanpepa]

 

BHZ is a bit like trainspotting. It has little relevance to reality but it enables people who have not enough to do to discuss its lack of relevance with other people who have not enough to do.

 

Alan

Sorry if you don't get it.

 

Pearl

 

Don't be sorry Pearl, each to their own as they say.

[John Mitchell]

 

 

BHZ is a bit like trainspotting. It has little relevance to reality but it enables people who have not enough to do to discuss its lack of relevance with other people who have not enough to do.

 

Alan

Sorry if you don't get it.

 

Pearl

 

Don't be sorry Pearl, each to their own as they say.

 

 

Yes, I sure wish that I had more to do.

[Inchiquin]

 

 

BHZ is a bit like trainspotting. It has little relevance to reality but it enables people who have not enough to do to discuss its lack of relevance with other people who have not enough to do.

Ah Alan, I would beg to differ. The recent downgrading in my BHZ placing is proving to be a good indicator of my placing in 'real life' test searches, which I also check from time to time. In these I was routinely in the upper quartile of results, and now I find myself in the lower quartile.

But you've hit my nail firmly on the head there when you talk about real-life searches. Surely that's what really matters? BHZ only works as a reliable indicator if ( a ) every contributor uses it and ( b ) they follow the rules and only have one image marked. But neither condition is, or is likely to be, true. I'm interested in where my pictures come in customer searches where every image found is competing directly against mine. I have a small set of sample search words and phrases which I check after every re-rank. During the period when my BHZ image steadily rose from page 13 to page 2, my position in many of my 'real-life' searches changed very little.

 

You would get a much better idea of your rank against a much larger sample of contributors by doing a search for, say, 'britain' or 'usa' than for 'bhz'.

 

Alan

[wiskerke]

 

 

BHZ is a bit like trainspotting. It has little relevance to reality but it enables people who have not enough to do to discuss its lack of relevance with other people who have not enough to do.

Ah Alan, I would beg to differ. The recent downgrading in my BHZ placing is proving to be a good indicator of my placing in 'real life' test searches, which I also check from time to time. In these I was routinely in the upper quartile of results, and now I find myself in the lower quartile.

But you've hit my nail firmly on the head there when you talk about real-life searches. Surely that's what really matters? BHZ only works as a reliable indicator if ( a ) every contributor uses it and ( b ) they follow the rules and only have one image marked. But neither condition is, or is likely to be, true. I'm interested in where my pictures come in customer searches where every image found is competing directly against mine. I have a small set of sample search words and phrases which I check after every re-rank. During the period when my BHZ image steadily rose from page 13 to page 2, my position in many of my 'real-life' searches changed very little.

 

You would get a much better idea of your rank against a much larger sample of contributors by doing a search for, say, 'britain' or 'usa' than for 'bhz'.

 

Alan

 

 

I don't get it. What's wrong with BHZ?

The other real searches are already there by the zillions?

 

So if you can see your real rank with a search for Britain, or Panda Bears in China no problem: use that one.

 

BHZ is not about your real rank, if we define Your Alamyrank as your real rank, or the real rank of a particular pseudo.

It's about your relative rank. It's about what your pseudo is doing over time. And probably the most important: how it responds to changes in your real rank: the dreaded or anticipated re-rank.

Or when something new is being introduced.

And it provides a relatively stable environment for testing your keywording; the choice of RM vs RF; ranking of your pseudos; creating a new pseudo and so on and so forth.

Basically answering the question what is happening when I do this?

All without any repercussion on your real rank or your CTR or your sales. Which would suffer if you start experimenting with real keywords and real pseudo's. It is like working with a real life model.

 

On the most basic level it tells me what to shoot or what not to upload because it will drag my pseudo down.

Or what to put into a rubbish bin pseudo. Or should I? Hmm maybe having a large rubbish bin pseudo is hurting my real rank real badly. I'd better find out. Back to the BHZ testing board.  Now how would I go about testing that?

 

wim

[Inchiquin]

I don't get it. What's wrong with BHZ?

 

 

Nothing wrong with it. But which is more useful, a search that tells you your relative rank against a small subset of contributors, or a search that does the same for a much larger subset, using subjects that you actually have in your collection?

 

It might take a little longer to find your image in a more wide-reaching search, but not significantly so because of the distribution algorithm. I did once start writing a script that would find the position of your first image in a search automatically without having to wade through hundreds of photos, but then Alamy changed the system so that the pseudonym was no longer matched in a standard search and I never found the time to re-design it.

 

Alan

[wiskerke]

 

I don't get it. What's wrong with BHZ?

 

 

Nothing wrong with it. But which is more useful, a search that tells you your relative rank against a small subset of contributors, or a search that does the same for a much larger subset, using subjects that you actually have in your collection?

 

It might take a little longer to find your image in a more wide-reaching search, but not significantly so because of the distribution algorithm. I did once start writing a script that would find the position of your first image in a search automatically without having to wade through hundreds of photos, but then Alamy changed the system so that the pseudonym was no longer matched in a standard search and I never found the time to re-design it.

 

Alan

 

 

Mark solved ;-)

 

wim

[NielsVK]

I've been trying to predict whether my pseudos' rank will move up or down at the next re-ranking by looking at some basic stats for the previous 12 months.

Of course actual rank is relative to other contributors so one pseudo's stats are not the complete picture but the trend is mostly obvious. Especially for my secondary pseudo my simple formula has been pretty accurate.

The blue lines below indicate a pseudo's position in the BHZ search results. The red lines indicate the revenue generated by the pseudo in relation to the number of views over the last 12 months. Both normalized to a number between 0 and 10 (higher means better rank).

These graphs don't consider zooms, I think they are relatively insignificant. Same for collection size.

 

 

I like playing with stats and this is just an oversimplified observation, but perhaps it helps those who have been repeatedly surprised/disappointed by the re-ranks

[Johnnie5]

OK, here is a real life search result for all you doubters.  I was demoted from page 12 to page 25.  A few weeks ago I did a search using the search terms "California tourism" and my images would show up on page one or two.  Some of you may say why use such unspecific search terms.  My and a few others theory is that photo researchers don't always know exactly what they are looking for.  They are looking for something that inspires them.  More general search terms may be useful in leading customers to our images.   When I use the search terms California tourism i get 63,616 images returned.  That is roughly 531 pages at 120 images per page.  I have 1857 images that use that combination of keywords.  Now my first image shows up on page 57 of 531 pages.  My second image shows up on page 76 of 531 pages.  I am sure that A customer is going to look through 6840 or so photos to find my first one.  The next question is what is wrong with Alamy's so called randomization algorithm?  For that matter the whole ranking system?  Ranking makes a difference.

[wiskerke]

I've been trying to predict whether my pseudos' rank will move up or down at the next re-ranking by looking at some basic stats for the previous 12 months.

Of course actual rank is relative to other contributors so one pseudo's stats are not the complete picture but the trend is mostly obvious. Especially for my secondary pseudo my simple formula has been pretty accurate.

The blue lines below indicate a pseudo's position in the BHZ search results. The red lines indicate the revenue generated by the pseudo in relation to the number of views over the last 12 months. Both normalized to a number between 0 and 10 (higher means better rank).

These graphs don't consider zooms, I think they are relatively insignificant. Same for collection size.

 

 

I like playing with stats and this is just an oversimplified observation, but perhaps it helps those who have been repeatedly surprised/disappointed by the re-ranks

That is an elegant graph.

Maybe repeat if possible with a 300 days period?

 

The problem with zooms is that a really bad pseudo has no zooms, and a really good pseudo has a low zoom/sales ratio. (If clients zoom an image they tend to buy it.)

While in between there are pseudos with all kind of reasons for a higher zoom number: cute cats; lovely ladies and the so called wtf zooms. It would be really difficult to filter all that out.

 

wim

 

edit: typo

[NielsVK]

Ah yes it's 300 days, not a year, thanks for that Wim!

 

I had a quick look and the global picture remains similar when I use 10 months. I did notice an interesting difference after changing from 12 months to 10 months though. My main pseudo had an exceptionally good October 2013. This boosted its rank significantly for the next 10 months. Last August the 'predicted rank' fell off the cliff as October 2013 was no longer relevant. Had there been a re-rank in September then my main pseudo would most likely have taken a hit. However as it happens October 2014 turned out to be a very good month as well which brought the 'predicted rank' back where it was, and the recent re-rank indeed did not change that pseudo's rank much. So re-rank timing can be (un)fortunate, perhaps this explains some of the unexpected results for some people?

 

Have a good weekend!

 

 

[wiskerke]

Ah yes it's 300 days, not a year, thanks for that Wim!

 

I had a quick look and the global picture remains similar when I use 10 months. I did notice an interesting difference after changing from 12 months to 10 months though. My main pseudo had an exceptionally good October 2013. This boosted its rank significantly for the next 10 months. Last August the 'predicted rank' fell off the cliff as October 2013 was no longer relevant. Had there been a re-rank in September then my main pseudo would most likely have taken a hit. However as it happens October 2014 turned out to be a very good month as well which brought the 'predicted rank' back where it was, and the recent re-rank indeed did not change that pseudo's rank much. So re-rank timing can be (un)fortunate, perhaps this explains some of the unexpected results for some people?

 

Have a good weekend!

 

Maybe 300 days makes it even more accurate.

Have a good weekend too!

 

wim

[NielsVK]

This last re-rank seems to have thrown into the game the spectre of negative or sub-zero ranking. I have a couple a pseudo's with lots of sales which are being trumped easily in the ranking by pseudo's with zero sales in the past year and one or two with zero zooms. Both in real life searches and BHZ. This leads me to the conclusion that whilst the rank of the zero sales/zooms pseudo's may be zero or just less, those with a lower ranking are getting much less than zero. 

 

I think it says somewhere in the Alamy info that if a pseudo has insufficient data to calculate an actual rank it will be given median rank. So yes a pseudo with zero sales and zero zooms can have a higher rank than a pseudo with sales.

 

In the graphs I posted you can see that both my pseudos have experienced sudden free-falls down to around 3. The 10-months periods preceding those 'dips' actually where very poor in terms of sales (long droughts) so I think median rank was assigned because the number of data points was below the threshold. After sales pick up eventually the pseudos go back to having an actual rank.

 

As my collection has grown sales have become regular so the 'dips' should hopefully not occur anymore

[Martin P Wilson]

 

 

This last re-rank seems to have thrown into the game the spectre of negative or sub-zero ranking. I have a couple a pseudo's with lots of sales which are being trumped easily in the ranking by pseudo's with zero sales in the past year and one or two with zero zooms. Both in real life searches and BHZ. This leads me to the conclusion that whilst the rank of the zero sales/zooms pseudo's may be zero or just less, those with a lower ranking are getting much less than zero. 

 

I think it says somewhere in the Alamy info that if a pseudo has insufficient data to calculate an actual rank it will be given median rank. So yes a pseudo with zero sales and zero zooms can have a higher rank than a pseudo with sales.

 

 

That would tally. It means that a median rank is about page 10/11 on BHZ.

 

 

Which is what is happenoing to mine:(

[wiskerke]

 

 

 

This last re-rank seems to have thrown into the game the spectre of negative or sub-zero ranking. I have a couple a pseudo's with lots of sales which are being trumped easily in the ranking by pseudo's with zero sales in the past year and one or two with zero zooms. Both in real life searches and BHZ. This leads me to the conclusion that whilst the rank of the zero sales/zooms pseudo's may be zero or just less, those with a lower ranking are getting much less than zero. 

 

I think it says somewhere in the Alamy info that if a pseudo has insufficient data to calculate an actual rank it will be given median rank. So yes a pseudo with zero sales and zero zooms can have a higher rank than a pseudo with sales.

 

 

That would tally. It means that a median rank is about page 10/11 on BHZ.

 

 

Which is what is happenoing to mine:(

 

 

Which may also mean that when you do pick up a couple of sales the system suddenly realises that your pseudo isn't actually doing all that well and sends you down to the dungeons of page 25.

 

wim

[wiskerke]

Ah yes it's 300 days, not a year, thanks for that Wim!

 

I had a quick look and the global picture remains similar when I use 10 months. I did notice an interesting difference after changing from 12 months to 10 months though. My main pseudo had an exceptionally good October 2013. This boosted its rank significantly for the next 10 months. Last August the 'predicted rank' fell off the cliff as October 2013 was no longer relevant. Had there been a re-rank in September then my main pseudo would most likely have taken a hit. However as it happens October 2014 turned out to be a very good month as well which brought the 'predicted rank' back where it was, and the recent re-rank indeed did not change that pseudo's rank much. So re-rank timing can be (un)fortunate, perhaps this explains some of the unexpected results for some people?

 

Have a good weekend!

 

For revenue, do you take Date Cleared; Date of invoice or Date Paid? It could make a difference, certainly with fluctuating revenue for smaller collections or pseudos.

With your one time really high sale, it's probably easy to see which one gives the most accurate results.

 

I've always thought the size of the collection or pseudo was a big factor. That may have been a fallacy, as it is actually only just views.

 

wim

[NielsVK]

For revenue, do you take Date Cleared; Date of invoice or Date Paid? It could make a difference, certainly with fluctuating revenue for smaller collections or pseudos.

With your one time really high sale, it's probably easy to see which one gives the most accurate results.

 

I use Date of invoice, I haven't recorded the other ones in my spreadsheet, I'd have to dig up the data from My Alamy. It seems accurate enough though.

 

I played with the data and the graphs a bit more and realized a few things.

 

Looking at where the median rank seems to be, and where my pseudos sit whenever they have enough data to calculate an actual rank, it seems that there must be a huge number of contributors with very poor performing collections (very low revenue to views ratio). It doesn't take much revenue even for my secondary pseudo to achieve a top 20% rank. However within that top 20% it seems much harder to move up. This is where the money is made and specifically by a small number of contributors achieving much higher revenues than most of us. So it's easy to move up from the bottom but the closer you get to the top the harder it gets.

 

For a lack of data on other contributors a logarithmic scale could be applied to my "BHZ prediction" to simulate this effect. Had a quick look and the resulting graphs seems to be much more accurate, ignoring the "median rank dips" and following the actual rank much more closely.

[wiskerke]

 

For revenue, do you take Date Cleared; Date of invoice or Date Paid? It could make a difference, certainly with fluctuating revenue for smaller collections or pseudos.

With your one time really high sale, it's probably easy to see which one gives the most accurate results.

 

I use Date of invoice, I haven't recorded the other ones in my spreadsheet, I'd have to dig up the data from My Alamy. It seems accurate enough though.

 

I played with the data and the graphs a bit more and realized a few things.

 

Looking at where the median rank seems to be, and where my pseudos sit whenever they have enough data to calculate an actual rank, it seems that there must be a huge number of contributors with very poor performing collections (very low revenue to views ratio). It doesn't take much revenue even for my secondary pseudo to achieve a top 20% rank. However within that top 20% it seems much harder to move up. This is where the money is made and specifically by a small number of contributors achieving much higher revenues than most of us. So it's easy to move up from the bottom but the closer you get to the top the harder it gets.

 

For a lack of data on other contributors a logarithmic scale could be applied to my "BHZ prediction" to simulate this effect. Had a quick look and the resulting graphs seems to be much more accurate, ignoring the "median rank dips" and following the actual rank much more closely.

 

 

With that single big sale, it would be easy to check: you only need the dates of that one. The effect must have been dramatic.

Oops: Unless the dates are very close together and all fall well within the 100 day period between re-ranks. (slaps forehead)

Well within: because we don't know if what we see as a re-rank is a real time process or the report of a process that has happened before. Or a real time thing with older data: something must tell the system what the data are. It's clearly not an automated task. Why would that be?

 

wim

[NielsVK]

 

 

For revenue, do you take Date Cleared; Date of invoice or Date Paid? It could make a difference, certainly with fluctuating revenue for smaller collections or pseudos.

With your one time really high sale, it's probably easy to see which one gives the most accurate results.

 

I use Date of invoice, I haven't recorded the other ones in my spreadsheet, I'd have to dig up the data from My Alamy. It seems accurate enough though.

 

With that single big sale, it would be easy to check: you only need the dates of that one. The effect must have been dramatic.

Oops: Unless the dates are very close together and all fall well within the 100 day period between re-ranks. (slaps forehead)

Well within: because we don't know if what we see as a re-rank is a real time process or the report of a process that has happened before. Or a real time thing with older data: something must tell the system what the data are. It's clearly not an automated task. Why would that be?

 

 

It was not a single big sale, there was one month with a relatively high total revenue. I only keep totals per month in my spreadsheet, not single sales.

 

My impression is that Alamy's software architecture is not particularly sophisticated or cutting edge. If you look at how modern web sites manage, analyze and use huge amounts of data about their visitors to generate profiles for them and serve them targeted content (all in real-time) then what Alamy is doing does not seem all that complicated. On the other hand real-time ranking might not really have a benefit, in fact it could mean that the same search can have a different result the next day which is confusing to buyers (now only happens around re-rank time).

[Reimar]

 

 

This last re-rank seems to have thrown into the game the spectre of negative or sub-zero ranking. I have a couple a pseudo's with lots of sales which are being trumped easily in the ranking by pseudo's with zero sales in the past year and one or two with zero zooms. Both in real life searches and BHZ. This leads me to the conclusion that whilst the rank of the zero sales/zooms pseudo's may be zero or just less, those with a lower ranking are getting much less than zero. 

 

I think it says somewhere in the Alamy info that if a pseudo has insufficient data to calculate an actual rank it will be given median rank. So yes a pseudo with zero sales and zero zooms can have a higher rank than a pseudo with sales.

 

 

That would tally. It means that a median rank is about page 10/11 on BHZ.

 

 Not sure about that.  I think a new pseudo lands about page 20. 

Oddly, my worst performing pseudo (I mean really bad), landed on page 17 this last time out, whereas it is usually a few pictures off last place.

[winstainforth]

Top of Page to so all fine. 

 

I've recently made a major change in the types of images I submit so will be interesting to see if that affects things in the long run.