Alamy stock photography

AlamySearch in detail

This article describes in depth how Alamy determines the order in which to display images in search results, planned future developments and gives some general guidance on how contributors can get the most out of their images on the Alamy site. Please note that all the systems discussed below are undergoing a process of continuous improvement. These notes may be updated from time to time to reflect any changes.

For the purposes of AlamySearch the following definitions apply:

AlamyRank: The score assigned to a collection based on the number of times images from a set of images have been clicked or purchased in proportion to the number of times they have been viewed by our customers.
AlamySearch: The system composed of AlamyRank + Search Engine + Diversity Algorithm.
Click: The click of a thumbnail by a customer to zoom the image.
Collection: Any group of images assigned the same pseudonym.
Combined score: The score that is calculated from AlamyRank + Search Engine + Diversity Algorithm.
Customer: Any registered individual client who has spent over a certain threshold with Alamy.
Diversity Algorithm: The mechanism to ensure that no single pseudonym dominates the results. It disperses the images evenly among suppliers of similar AlamyRank and relevancy.
Pseudonym: The name used to group a collection of images. A contributor can have one or more pseudonyms associated with their account. This is to allow for those contributors who represent multiple photographers, or for those who wish to segment their collections to reflect one or more areas of specialisation in their work.
Sale: Any image purchase by a customer who has spent over a certain threshold with Alamy.
Search engine: The system that identifies which images on Alamy contain the words searched for by a client. These words may be in any of the caption, keyword or description fields, or pseudonym/agency name.
View: The thumbnail of an image being presented in search results to a customer (on a page that is viewed by that customer).

Background

Alamy’s policy of not editing images, charging a low commission, and leaving the keywording to its contributors has proved popular with customers and contributors. Alamy also made a strategic decision never to own an image, so it can avoid any conflict of interest that may encourage other agencies to favour images from the collections that return the highest percentage.

Alamy has always faced the challenge of delivering the right images for a search. Since launch, the sort order had a built-in bias that favoured agencies over photographers. This was the result of a search order logic which gave an approximately fair share of views between contributors. The ‘fairness’ of this system was eclipsed over time by a growing number of images submitted by individual photographers who were given an artificially low share of customer views. The result was a lottery where it became a matter of chance which images were presented to a client.

Alamy appreciated the limitations of this system and began investing heavily in a new approach to create an environment where all contributors could compete on an equal footing.

The first version of the new search system AlamySearch, was launched on 10th October 2006. A key component is AlamyRank, a landmark in the industry in terms of fairness of presentation of images to clients. Where other image libraries may order images using solely one parameter, such as reference number or date uploaded, Alamy’s new system relies exclusively on client activity to objectively place the most popular image suppliers higher up the sort order.

What is AlamySearch?

AlamySearch is a new system composed of three elements: AlamyRank, the Search Engine and the Diversity Algorithm. These elements combined determine the positioning of a collection of images within a set of search results and create a "Combined Score" for a collection.

AlamyRank

AlamyRank assigns a score to a pseudonym based on the number of times images from that collection have been clicked or purchased in proportion to the number of times they have been viewed by customers. The result is an AlamyRank per collection (pseudonym).

The simplified formula for this relationship is:

An image of the AlamyRank formula

The Search Engine

The search engine finds images that match the search terms entered by the customer. If a customer searches for ‘dog’ then all the images containing the word ‘dog’ will be returned. Currently, these words may be in any of the caption, keyword or description fields, or in the pseudonym/agency name. It also deals with pluralisation (e.g. ‘dog’ also finds ‘dogs’) and stemming (e.g. ‘think’ also finds ‘thinking’). Alamy’s current search engine uses third party software which cannot be fully tailored to Alamy’s needs. It has been altered slightly for the implementation of AlamyRank to force the system to display RM and RF images to reflect the proportion of each licence type, which is 1/3 RF, 2/3 RM on the site. This means that for every two RM images in the results one RF image is shown.

In 2007, Alamy aims to launch the first version of its new search engine that will allow better control over how the system behaves, particularly in terms of relevancy. The current search engine on Alamy gives equal weighting to caption, keywords, descriptions and pseudonyms, which is not ideal. The final plan is still under review; however some of the likely changes in the new system are as follows:

Reduce the significance of the description field relative to the caption and keyword fields. This will overcome the problem of contributors using the description field to provide background information for images or photo journalistic descriptions. While useful to some buyers, it can be a hindrance to searching.
Reduce the significance of the pseudonym field relative to the caption and keyword fields. This will ensure that, for example, when a client does a search for "fish", a contributor called David Fish does not have his images returned before images of fish (which contain the word fish in their caption or keyword fields).
Spam prevention - repeating words or phrases won't increase the relevancy score.
Proximity - words in sequence that match phrases entered by customers will carry more weight than words not in sequence For example, if a client searches for "Blue Whale" an image with the words "Blue" and "Whale" next to each other in the caption or keywords will have a greater relevancy score than an image containing the words "Blue, Sea, Humpback Whale" in this order.
Date - ability for customers to order the search results by date added.

The Diversity Algorithm

The Diversity Algorithm comes into play once AlamyRank and the search engine have established which images have the highest score. Its purpose is to ensure that the images returned are inter-dispersed amongst images of similar rank. This means that search results are not dominated by the collections with the highest combined score and that no single pseudonym dominates the results when others are available. The strength of the Diversity Algorithm diminishes if the gap between the collection rankings is significant. This means that clusters of images from a single collection can appear close together in the search results if there are no competing images available of a similar AlamyRank and/or relevancy score. Given the large number of images on Alamy it is unusual to see this effect on the first few pages of results for most searches.

How does AlamyRank work?

AlamyRank is driven exclusively by customer activity. There is no manual intervention in the system. Various measures are used to calculate the AlamyRank for a pseudonym.

What is measured

The decision on what types of client activity to measure in this first version of AlamyRank was based on a detailed analysis of the available data alongside Alamy’s understanding of the purchasing patterns and activity of customers on the site.

In this version of AlamyRank only sales, clicks and views are used to calculate the AlamyRank for a collection. Sales were chosen as they are a clear indication of a client’s preference for an image, although the monetary value of a sale is disregarded. Clicks were chosen as a strong correlation exists between a click (zoom) and a purchase. In the nine month measurement period there was, on average, a purchase for every 20 clicks. The correlation between both shopping carts and lightboxes in relation to a sale was, however, not statistically significant so this type of activity is not used in calculating AlamyRank in this version.

How does AlamyRank determine a rank for a pseudonym?

The current version of AlamyRank produces a rank per pseudonym and not per image. If a contributor's images are viewed frequently but not clicked or purchased very often, AlamyRank interprets this to mean that the contributor's images are less popular than a contributor whose images are clicked or sold more often in proportion to how many times they have been viewed. A lower AlamyRank would therefore be assigned to this pseudonym. The ultimate aim is to develop a search system that produces a rank per image.

When AlamyRank went live the ranks for all collections had been calculated on nine months of customer activity. When it is next calculated, the ranks will be updated to reflect any changes.

All collections that have recorded a sufficient number of views to be statistically significant have had their true AlamyRank applied. All collections that have insufficient views (these might be collections added near to the end of the measurement period, or new collections, etc) have the median AlamyRank applied. Applying the median score to new collections should give them enough visibility from which to calculate a true AlamyRank. Some ranks will go up, some will stay the same, and others will go down.

AlamyRank ignores whether a collection of images is from a single photographer, a specialist agency, an RF agency, a national collection or a small cooperative of photographers. It is only concerned with how many times images within a collection have been viewed, clicked, and purchased.

How does AlamyRank affect contributors?

Alamy’s hypothesis is that, whilst the order in which images are displayed is important, the absolute position of an image is not particularly significant in determining which image will be purchased. AlamyRank in its present state should not influence clients' purchasing patterns as they generally look through many pages of images before buying. However client purchasing patterns may change over time as AlamyRank evolves.

AlamyRank in its current state is already a big change and there will be winners and losers. For most individual photographers AlamyRank is unquestionably a change for the better. The old system considered all contributing photographers to be part of a single Agency that we titled "Alamy photographers". These contributors had a disproportionately small share of the customer views relative to agencies. Search results are now divided up between pseudonyms. The result is that whereas in the month before AlamyRank went live, individual photographers (who represent 60% of the images on the site) were getting 40% of customer views, they now get 60% of the views.

AlamyRank rewards contributors who consistently keyword their images accurately and, over time, will penalise those who don’t. It also penalises contributors who submit too many similar images. Whilst the system has no way of identifying “similars”, supplying many similar images results in a contributor’s images being viewed more frequently by customers. A lot of views can only be good if there are enough clicks or purchases to be competitive with other contributors with a similar number of views.

There will, however, be some contributors who do not benefit as much as they deserve to. By creating an average combined score, AlamyRank may penalise contributors for activities which were in their best interests in the past. An example is where contributors have used the description field to provide background information for images or photo journalistic descriptions. In AlamyRank this will increase the number of times images from a collection are viewed, which will not benefit a contributor unless the number of sales and/or clicks is also high. One solution to this, as mentioned above, is to reduce the significance of the description field relative to the caption and keyword fields. This is likely to happen in a future version of the search engine.

AlamyRank has no way of determining which image in a particular category is the best. For example, it does not know which is a contributor’s best dog image. All images under one pseudonym have the same AlamyRank assigned. When a client performs a search, the search engine first selects images which contain the required search terms in the caption, keyword and description fields. It then displays these according to the AlamyRank for the image's pseudonym. If there is more than one image with the same keywords and caption and the same pseudonym, it will return these in a random order. Finally, the Diversity Algorithm disperses images of a similar rank in order than no one supplier dominates the results.

What can contributors do to improve their AlamyRank?

Alamy has released Version 1 of AlamyRank and will be releasing new versions in time. However, there are things contributors can do to make the most of the new search system.

Keywording

Alamy advises that contributors check their keywords carefully to ensure that they are both spelt correctly and are relevant to the image concerned. Keywording both literally and conceptually is still advisable, but contributors should be careful that their captions and keywords are entirely relevant to the image. It is useful for contributors to put themselves in the buyer’s position to see if they would be interested in an image for a particular search term. As a simple example, if keywording an image of a cat it would be best not to add a caption saying ‘Cat which is afraid of dogs’. Each time this image is returned on a search for ‘dog’ it will be viewed by a client and, in the majority of cases, not clicked or purchased. This will negatively affect the AlamyRank for the pseudonym that is associated with this image. Finally, if the plural and singular applies to your image, then contributors should add both.

Editing

Alamy has always stressed the importance of editing images before submitting them. For a series of images, it is best to submit a portrait, a landscape and an obscure/conceptual angle from each shoot/subject area to reduce image redundancy and improve the visual quality of results when a client performs a search. Supplying too many similar images will be off-putting to clients and may result in a lower AlamyRank as described in How does AlamyRank affect contributors?

Using pseudonyms to group images

Pseudonyms allow for those contributors who represent multiple photographers or for those who wish to segment their collections to reflect one or more areas of specialisation in their work. As an AlamyRank is assigned to a pseudonym, the rank is the result of client activity across a collection of images. An individual photographer may therefore have had more than one AlamyRank assigned if they have more than one pseudonym.

If all images are grouped into one pseudonym this could lower the overall rank. If a contributor is an excellent wildlife photographer but their travel imagery is not as strong as the competition, it may be advisable to split their collection into two pseudonyms. The reason for this is that the comparatively low number of sales and clicks in proportion to views for the travel images may bring down the overall AlamyRank.

Future plans

The first release of AlamyRank is already a huge improvement on Alamy’s old searching system. The only way to objectively measure the impact of AlamyRank is to build a realistic picture of what is happening in terms of image placement and sales over time. The aim is to recalculate AlamyRank on a regular basis at time intervals to be determined from what is learnt over the first launch period.

To ensure that customers and contributors benefit fully from the new system, feedback will be provided to contributors to help them improve the rank of collections. The goal is to allow contributors to make informed decisions about how to submit and describe their pictures. AlamyRank will become more interactive in the future and will include tools that report customer activity and suggest modifications to submissions and metadata.

Concluding remarks

Alamy has achieved its first objective of building a large volume of imagery from a wide variety of sources. There is now a level playing field so that contributors can compete effectively in the marketplace.

The second objective is to create a more efficient and dynamic marketplace for commercial photographs, where contributors can maximise their sales potential on Alamy. To achieve this Alamy will be developing tools to allow images to be annotated more easily in response to client activity.