eBay faucets pc imaginative and prescient to rework on-line buying

All of the classes from Rework 2021 can be found on-demand now. Watch now.

When eBay rolled out image-based seek for buyers practically 4 years in the past, the corporate was among the many first to deploy pc imaginative and prescient expertise in ecommerce. eBay has realized many classes since then, and it’s now engaged on bringing that innovation to sellers on its platform.

In a dialog at VentureBeat’s Rework 2021 digital convention, eBay’s new chief AI officer, Nitzan Mekel-Bobrov, shared some insights with Maribel Lopez, founder and principal analyst at Lopez Analysis. Mekel-Bobrov mentioned the challenges of creating image-based search expertise, the important components of any AI technique, and what’s subsequent in pc imaginative and prescient.

The challenges of image-based search

Providing consumers the power to search for related or precise gadgets utilizing photographs as a substitute of textual content has grow to be pretty commonplace within the trade, however launching the function in 2017 was a giant deal, Mekel-Bobrov mentioned.

Knowledge offered a few of the hardest obstacles, particularly with the vary of things eBay is understood for.

“The adage that knowledge is king is true, most likely nowhere greater than in coaching CD fashions for item-matching,” Mekel-Bobrov mentioned. “And for us, this was compounded on the time [by] the truth that we don’t management all of the methods through which our sellers {photograph} their gadgets.”

{Hardware} additionally posed points within the first years, Mekel-Bobrov mentioned, and the corporate needed to discover inventive options.

“First, we did our coaching individually,” he defined. “We skilled the fashions on the general public cloud, the place we might entry [graphic processing units]. Then we did the inference and deployment regionally. We additionally arrange a collection of what had been primarily stateless programs, holding the whole index in reminiscence.”

eBay was additionally one of many early customers of product quantization. That decreased the dimensionality and measurement of the issue throughout search.

Organizational challenges stay eBay’s greatest enterprise. Launching a product that requires many various groups with “super” upstream and downstream dependencies presents myriad points, Mekel-Bobrov mentioned. “A whole lot of the workflows wanted remodeling to permit for integration of pc imaginative and prescient capabilities.”

Moreover, ecommerce class taxonomy modifications primarily based in the marketplace and stock. “Because the taxonomy innovation progresses, we have now to maintain up with it on the picture search web site and constantly adapt and evolve not simply our fashions however our total strategy generally,” Mekel-Bobrov mentioned.

The important components of AI technique

Having solely been with the corporate for 2 months, Mekel-Bobrov mentioned he’s been spending most of his days occupied with eBay’s AI technique. His plan focuses on three broad classes: taking a customer-first strategy, democratizing AI amongst groups, and nailing down the information infrastructure.

The query ought to by no means be what new expertise may be adopted, however what buyer issues may be solved, Mekel-Bobrov mentioned. “For us, as a two-sided market … the mediation of that relationship — there’s a ton of alternative and complexity there that AI can actually remodel.”

The quickly altering nature of AI additionally requires companywide upskilling and coaching. “We wish to put these applied sciences into the arms of builders far more broadly than AI specialists, even product managers and analysts,” Mekel-Bobrov mentioned. “The query is, how will we set up ourselves such that we are able to have a smaller cohort of AI specialists construct differentiated core capabilities that may then allow groups throughout the corporate to leverage these and construct distinctive merchandise in their very own domains?”

Lastly, a key aspect of any AI technique is knowledge. Mekel-Bobrov mentioned an organization should ask the place the information is whereas offline and streaming, in addition to the way it’s collected, labeled, managed, and searched. “The extra we’re doing at scale, the extra we’re doing with actual time, the extra we’re doing with dynamic studying and reinforcement studying, the extra we have now to push the boundaries of those infrastructure issues,” Mekel-Bobrov added.

What’s subsequent?

The ecommerce large is now “turbocharging” its platform to make pc imaginative and prescient an integral a part of the expertise for each consumers and sellers, Mekel-Bobrov mentioned.

eBay just lately launched an image-scanning software on its app, beginning with buying and selling playing cards that sellers can scan with their telephone to auto-populate and create listings.

“If you concentrate on a person vendor or small enterprise that should constantly create new listings at a excessive quantity however needs to keep up the best high quality of data, promoting instruments primarily based on pc imaginative and prescient could be a actual game-changer,” Mekel-Bobrov mentioned. The final word purpose is to keep up belief between consumers and sellers.

Throughout the COVID-19 pandemic, eBay has used pc imaginative and prescient detection and classification fashions on gadgets like hand sanitizers and masks to determine listings with inflated costs and merchandise that make false well being claims.

Since November 2020, eBay has eliminated 50 million listings that violated its COVID-19 insurance policies, Mekel-Bobrov mentioned.

“We’re constantly making each effort to make sure that anybody who sells on our platform follows native legal guidelines, in addition to eBay’s insurance policies,” Mekel-Bobrov mentioned. “Doing that at a scale requires quite a lot of inventive pondering and superior applied sciences like pc imaginative and prescient.”


VentureBeat’s mission is to be a digital city sq. for technical decision-makers to realize information about transformative expertise and transact.

Our web site delivers important data on knowledge applied sciences and methods to information you as you lead your organizations. We invite you to grow to be a member of our neighborhood, to entry:

  • up-to-date data on the themes of curiosity to you
  • our newsletters
  • gated thought-leader content material and discounted entry to our prized occasions, similar to Rework 2021: Study Extra
  • networking options, and extra

Turn out to be a member