Did you miss a session from GamesBeat Summit Next 2022? All sessions are now available for viewing in our on-demand library. Click here to start watching.

Style matters a lot to Etsy’s millions of merchandisers. There’s roughly 43 different categories their products occupy, from knitted ware and cosmetics to midcentury modern decor, and the popularity of each style waxes and wanes depending on seasonality.

Typically, classifying items requires a great deal of manual labor. But Etsy’s researchers last year embarked on a project to detect styles automatically across the storefront’s over 60 million listings.

“[We wanted] to apply this to our buyers to try to get a sense of their style,” said Etsy CTO Mike Fisher during a session at VentureBeat’s Transform 2019 conference. “Tomorrow they might be looking for a wedding gift, and the next week, they might be looking for a gift for themselves because they want to celebrate a special occasion. We’re trying to understand each of those journeys and what style they’re after at those particular times.”

The researchers leveraged a computer vision system that took into account a vector of over 50,000 dimensions from a year’s worth of style data. Impressively, in experiments detailed in a recently published paper, it predicted listing styles with 76% accuracy and even managed to associate less obvious products — like images of whales — with appropriate themes (nautical, in that case).


Intelligent Security Summit

Learn the critical role of AI & ML in cybersecurity and industry specific case studies on December 8. Register for your free pass today.

Register Now

“A lot of the merchandisers currently have to do [categorization] manually, [so] we’re seeing if we can apply some of this [research] and do that almost in real time,” said Fisher.

It’s hard to believe now, but the cofounders of Etsy — Rob Kalin, Chris Maguire, and Haim Schoppik — operated out of a tiny Brooklyn apartment in the craft ecommerce shop’s formative months. Just two short years later, Etsy had grown to 450,000 registered sellers, and today, 2.1 million people hawk their wares to nearly 40 million buyers on the platform.

In the months following Etsy’s initial public offering in 2015, AI played a key role in its growth. In 2017, the company launched context-specific ranking and personalization tools, followed by a combination of sophisticated computer vision and natural language processing capabilities. Thanks to this robust and growing framework of AI tools, Etsy’s search engine today can surface results “most relevant” to buyers based on signals like the time of day and buying histories.

The AI-forward approach has driven an increase in gross merchandise sales by hundreds of millions of dollars over the past two years, according to Fisher. (Total gross merchandise sale sat at around $3.93 billion on the platform, as of 2018.) But it’s a constant evolution — Etsy uses a billion events a day that they to retrain its models, and it’s investing in research to mitigate unwanted bias that might crop up in its recommender systems.

Etsy’s machine learning architecture runs on Google Cloud Platform, and the firm continues to invest heavily in engineering. In May, Etsy opened a machine learning center in Toronto, following on the heels of locations in Brooklyn and San Francisco. And it counts 400 people among its engineering team and 30 in its data science team, the latter of which it says is roughly tripling each year.

VentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.