First, I explored how other sites incorporated the concept of image similarity (known as "More like this" in the industry) on their sites. The best way to get your creative juices going as a designer is to see what’s already working for your users - don’t try to reinvent the wheel unless your company is aiming for a signature moment. In this exploration, I wanted to learn how competitor sites defined image similarity.
I explored options on how we could present More Like This through good ol’ sketching and mocks to see what would make sense to our users. It was logical to place the feature on pages with image results (search, lightboxes, and galleries) because the user would want to find similar imagery after they've browsed their initial search results. From there, I explored how image similarity is defined by our customers. Was it just subject matter similarity, visual similarity, or both?
After reviewing with the greater team, we believed the best option provided the simplest experience: let users click on a More Like This link and then present both visually and semantically similar results in a More Like This results page. I placed the link on the image preview and product details page because these are locations in the search experience in which the user is interested in an image and may want to see more similar to it.
See functional specs
In collaboration with Development, we fine-tuned the More Like This algorithm in order to return both visually and semantically similar results, but found that both could not be served up at the same time due to slow performance caused by combining the visual and semantic search engines. This would hinder the experience, especially for our international customers with slower Internet speeds. We decided to rely on just an image’s metadata to define visual similarity. Subject matter relevancy is the number one search criteria for our customers. Visual similarity is secondary because our users can easily revise an image in-house to meet a project’s visual needs.
After its release, we found that more users opened an image's product details page in a More Like This search (71.9%) compared to regular searches (22.6%). The higher click-through activity indicates that More Like This provides a better selection of the right images for our customers.
In a later customer study, we discovered that customers do expect More Like This to return images of both visual and semantic similarity, though semantic similarity is still most important. As a future project, we should improve our search engines so that More Like This can return both visual and subject matter relevant results without deterring the site performance.