How does the search algorithm work?? Ideas!

By the way I know this is long and in-depth but I think its worth it
Some months ago while working on marketing for my images here on RB I though about the fact that RB say they advertise to bring people to the site but that it is up to us to bring the buyers to our page. It occurred to me that in searching for a particular subject, for instance “Droplets”, that buyers are only going to go through a certain number of pages of search results before they either decide on something or move on to some other search.
I asked a number of people to have a look at a RB search and tell me how many pages they would be willing to go through before moving on.. The results ranged from 5-10 pages. What this says to me is that in any given search your art works need to be ranking in the top 10 pages, that is the top 240 work for that search, to have a reasonable chance of being viewed and subsequently bought.
Now at time of writing this a search for Droplet produces 270 pages of results, thats approximately 6480 works.

So after finding out that anything, to have a reasonable chance of selling well, needed to rank in the top 10 pages of a search I tried to find out how search rankings worked I posted a forum post about it.
One of the answers I got “The order of what appears when you search is controlled by the RB popularity algorithm. The ingredients of the formula are secret but it is some kind of mix from the number of sales, the number of comments, the number of favorites and the number of views. Could be they added some other stuff lately as the results some times are very odd the last few months.” from Jo Beerens suggested that it was basically based on Views, Favs, Comments, and Sales

My research into how ranking algorithms work from outside RB also suggested that the various ingredients are given a weighting and that this weight combined creates the search order
The ingredients as best I can figure them out are:

  1. Tags on the image
  2. Sales (To actual buyers as opposed to sales to the artist)
  3. Favs

(Where sales to the artist fit in I’m not sure yet)

To test this I took my image Droplet on Bottlebrush and did a search for “droplet”, which as part of its name and also one of the tags used should come up in the search results, and then went through each page until I found my image. I then recorded its position by page number, column and row and then also recorded its sales, views,favs, and comments.

I then Favourited it and (after 24 hrs for the search to update) repeated the process, I continued this each day, marketing this particular image heavily on both RB and outside on the web in general, recording the results each time.
The results suggested that the weightings are in the order listed above with Tags being the heaviest and Views the lightest
At one point I decided to buy some of my image as cards to use as real world promotional material and this image was included. I bought 5 copies of this image as cards and I repeated the search and record position exercise I’d been doing expecting to see a big jump in where it ranked due to 5 sales. But it didn’t happen! Why? Well after much head scratching and pencil chewing I am starting to think that Sales to the artist I.e. where you buy your own work while logged in. are either not included in the ranking algorithm or are afforded such a low weighting that it makes next to no difference. In thinking about this some more it makes sense, If sales to the artist imporved rankings as much as “Real” sales then Artists could artificially improve their rankings by buying the artworks and selling them in the real world. Which is exactly what each artist wants to do.

Unfortunately I stopped recording the search results for my image before I made a sale for real and so have no data to compare rankings before and after the sale (if someone else out there cares to do such a experiment and post the results I would be eternally grateful)

So the BIG question is “what does all this mean for me?”
Well several things:

  1. First Tags are weighted I.e. Tags are given a total weight maximum of 100 at 10 total tags (this is only an example I don’t know the actual figures) so if you have 8 tags you have a total weight of 80 and a weight for each tag of 10. but if you have 20 tags and the maximum weight is limited to 100 then each tag is only worth 5. This is why you are better off doing a maximum of 10 very relevant tags than 10 very relevant and 40 vaguely relevant tags
  2. Second Sales to real customers are important (or what RB thinks are real customers) My suggestion is if you are going to buy your own work to sell in the real world then log out before buying because you still get the money from the sale (you are no more out of pocket than if you stay logged in) and it counts as a “real” sale to RB thereby increasing that image on the search rankings
  3. Third Favs are the third most weighted ingredient because people seem to only view stuff that interests them but only a portion of those people comment and only a portion of those again Fav a work I.e for a work to get fav’d it must be very liked by that person
  4. Fourth Comments weighted less than favs for the reason mentioned in the previous point
  5. Fifth Views weighted the least (unless sales to the artist are actually weighted very small but thats just conjecture)

So another way of explaining it might be to say:
(This is an example only the numbers are made up off the top of my head)
1000 views = 200 comments = 50 favs = 1 sale to a “Real” customer

I hope this helps you to understand how the system works and why a small amount of relevant tags are better
( I wonder how quickly I’ll get a Bmail from RB management asking me to shut this down for revealing their “Secret Formula” for search rankings??)

Update: 09/11/10 as of today there are 30592 PAGES of photography (Media tagged as photography that is) so thats 611840 induvidual pieces at 20 per page
Just thought you’d like to know


If you liked this please favourite this and spread the word
also please visit my Journal here about looking for work as a photographers assistant

Join my Facebook Fan page Here
Subscribe to my Blog posts here
or my Journal posts here

Journal Comments

  • Alixzandra
  • Richard Keech
  • Mark Tull
  • Richard Keech
  • Sonja Wells
  • Richard Keech
  • Garth Smith
  • Richard Keech
  • Garth Smith
  • Richard Keech
  • Cindy Schnackel
  • Richard Keech