Best Match 2.0 Summary

On 2008-12-09 15:09:33, bitter wrote:

Photographers love to hate Best Match, snickering at its name, BM– and discussing it at length in the shadows of forums everywhere. We know. The mystery of how BM functions is being analyzed regularly by everyone.

With BM, we have one, clear goal: to put the best content in front of our clients. We have finite slots on the first page and tens-of-thousands of files vying for those spots. We understand we won't make every artist happy, but we do our best to be fair when slotting files into BM search results.

A long time ago we developed an algorithm to rank keywords on each file. Since then, we've been tracking data for every single file on iStock. Guess what? It works. Starting in mid-January we will start applying these results to BM searches. For clients it will mean more accurate, more meaningful and more relevant results. For artists it means a massive shift. The results for everyone, are going to be very, very different.

This development is part of a multi-year strategy that we all undertook, starting with the CV (Controlled Vocabulary) that we took (and improved) from Getty Images, nearly three years ago. For those who are unfamiliar with the history, this is the technology that allows us to translate keywords in multiple languages automatically. Next, we all went through a year of disambiguation. You remember if you were here in 2007, the painful steps of going through each file and clarifying and exacting the keywords for every file in your portfolio. Most recently we began our ZTS (Zero Tolerance Spam) program, which is an ongoing initiative to ensure that there are no irrelevant terms or keywords in every file. It hasn't been easy and we understand the great amount of work everyone has endured to make our search results the best they can be. The quality of results has vastly improved since 2006 when we began this endeavor.

With the launch of our new keyword ranking system the culmination of years of work by thousands of people will come to fruition, providing iStock clients with what we think will be the most impressive and most exact search results, for community created media, found anywhere on the planet. In January when the new Best Match launches, it will finally live up to its name and accurately give clients the Best Match for their search terms. We will begin pulling leavers and turning dials to slowly turn up ranking in BM starting next week!

Posted By rogermexico:


Yeah, I know it's kind of a contradictory sounding message, but the goal is to have one last big shake so that we can move to something more stable. By blending this ranked keyword data into the algorithm, it should make it on the whole more stable. In the past we've been seeing really wild swings whenever the guys have gone into tweak the mix - this is aimed at alleviating those swings and giving us something more reliable.

Posted By kkthompson:



No. There will be slow & steady modifications as we cruise along.



Oh hell, I wanted to make you all sort every keyword for every file... [ducks] but I'd be an old man before it happened. Doing it algorithmically in the back-end was the next best thing. Can you imagine the beat-down in the forums if we'd have asked to do that? I shudder to think.

How does it work? Not really gonna give to many details, as we're pretty tired of people trying to game the system. You aren't necessarily penalized for bad keywords: they just don't return results.

Posted By bitter:


I'm happy you noticed how carefully I chose my words. We are not promising everyone will like the new BM, nor promising increased royalties to anyone or group in particular. Our goal is to improve the search experience for clients. That's all. We have to assume that the theory of well keyworded files being found and purchased more often than poorly keyworded files is true in this endeavor.

Posted By kkthompson:



Yes, this will happen in the New Year, so don't do anything to your keywords.

Posted By WendellFranks:
This is probably a really dumb question, but does the order in which we list keywords in a file influence the search? In other words, should the most important keywords be listed first? I haven't done that and got to wondering. Thanks!

Posted By Susan_Stewart:
While it's clear that Best Match 2.0 isn't fullly implemented yet, I'm not clear as to whether the keyword relevancy is included at all yet (ie whether the gradual prcess of including their effect has started). And if keyword relevancy is being included now, or this is the base search on which the keyword relevancy is going to be imposed why are undisambiguated old files appearing at the front of searches so much when this seems to fly in the face of using keyword relevancy?

Posted By Eulenblau:
Am I right understanding that titles and descriptions will be a factor in BM 2.0 and that I should edit them? It always seemed a bit odd to me to use the word "sailboat" in the title and in the description when the image clearly showed a sailboat and "sailboat" was a keyword anyway, so in the past I had the tendency to be a bit more creative in chosing titles and to be quite brief in my descriptions. Will that work against me now?

Posted By VLIET:







From which piece of text did you draw that conclusion? I haven't read a statement about titles & descriptions.

Posted By Eulenblau:
Am I right understanding that titles and descriptions will be a factor in BM 2.0 and that I should edit them? It always seemed a bit odd to me to use the word "sailboat" in the title and in the description when the image clearly showed a sailboat and "sailboat" was a keyword anyway, so in the past I had the tendency to be a bit more creative in chosing titles and to be quite brief in my descriptions. Will that work against me now?