Last week, eBay enhanced visual shopping on its mobile app. Users can now drag and drop images of an item into the eBay mobile app search bar to help them find such items for sale on eBay. Users had already been able to use Image Search to take a photo of an item and find it on the eBay app.
But how does eBay know which listings match the item featured in a photograph?
eBay Image Search utilizes AI (artificial intelligence). Earlier this year, eBay executive Scott Cutler outlined his thoughts on AI and said eBay was trying to use the technology to personalize shopping.
“AI is helping us organize the billions of listings, inventory and options into meaningful experiences, in how we connect buyers and sellers, what we offer, and how we interact. Advancements in machine learning, computer vision and natural language processing are helping create a shopping journey as individual as each customer.
“This is how AI gets personal. At eBay, it touches every aspect of our business, and we’re using it to customize an experience for each and every person using our platform. Our focus is on enabling an inspiring interaction, not a simple transaction.”
In last week’s post about Image Search enhancements, Seema Jethani, Product Lead for Search Platform and Venkat Medapati, Product Lead for Buyer Engagement & Discovery provided an example:
“Let’s say you’re looking for a new chair, so you type “egg chair” into the eBay search bar and you see a list of results for your search. If one chair in particular stands out, now you can search for more products that look like it—just drag and drop the image of the chair you like into the search bar to see visually similar results.”
Jethani and Medapati explained, “Behind the scenes, when you submit your image in the search field, the neural network converts your image into a vector representation. Then, the vector representation of the image that you submit is compared against more than 1.1 billion live listings in eBay’s marketplace, using nearest neighbor search. Finally, we surface the best-matched items, ranked by visual similarity.”
The two engineers said eBay had more than 2.1 billion images. “We had to train our models to cancel out the noise to better understand what you might be looking for and show you items that are visually similar. Over the past several months, we have also invested in training models specifically for visually rich categories such as Fashion and Home & Garden. We focus on images and text labels specific to those categories to return better quality results.”
The new visual shopping feature will be available starting in August on Android and iOS for the US, UK, Germany and Australia. The full post is found on the eBay corporate blog.