What is Deep Learning? How is it Affecting Marketing?
Talking about AI-driven advances in image search immediately brings us to the subject of deep learning. Its development is explained in the Nvidia blog as
Artificial Intelligence >> Machine Learning >> Deep Learning
 |
| Deep Learning Layers |
The post explains that what we can do at present falls into the category of “narrow (specific) AI”. Examples include image classification on Pinterest and facial recognition on Facebook. Machine learning uses massive neural networks, running huge amounts of data through a network until it learns to recognize the item with near perfection. The “layers” shown in the dog recognition graphic are at the heart of deep learning. Speech recognition and image recognition are both products of deep learning technology. The impact of voice technology was discussed in my recent post on voice search. Image recognition is, of course, the enabler of image search.
Image search is interesting in a general sense, but the recent presentation I developed on Social Commerce opened my eyes to some of the ways image search—fueled by AI—could offer new shopping opportunities. That presentation focused on Pinterest, clearly a leader in the social shopping space, and on Facebook whose sheer size makes it impossible to overlook its initiatives in social commerce.
The clear leaders in the image search space are, not surprisingly, the leaders in voice technology—Amazon, Apple, Google and Microsoft. You could probably add Chinese web services firm Baidu to that list, but less is known about the current status of its AI.
Amazon has been in the space for a considerable time, using images as one input in its recommendation engine. Over time, it has added AI/deep learning to its use of images. Amazon recently made news when it opened its deep learning framework DSSTNY to other developers as an open source platform. There was speculation in the trade that Amazon was using this as one way to catch up with the technology of its more recent rivals, especially Google. Facebook also has made its facial recognition software open source. Facebook’s facial recognition services have been a special target for privacy advocates, a subject in itself. Apple is using facial recognition to unlock the iPhone X.
Google is making advances in deep learning AI research on several fronts. In 2011 it introduced Rank Brain (discussed in Chapter 10) as part of its search process. At present, it also uses deep learning in a number of ways:
• To help categorize images on the web to improve search results. It also allows image enhancement, essentially filling in the blank, when images are missing detail.
• Google Cloud Video Intelligence allows videos to be segmented and analyzed for content and context with automated summaries provided. It can be used to search for various types of meaning including suspicious content. See a video demo on YouTube.
• Language recognition technology is essential to Google’s growing line of home assistant devices.
• Google also uses deep learning to improve recommendations on YouTube, thereby keeping users on the site longer.
• It is enhancing Google Image Search with Pinterest-like suggestions about related content. For instance, it will highlight when a recipe is available for a particular food image.
Remember—it’s still early days for these technologies. Nevertheless, there are already useful commercial applications. The fashion industry with all its visual content provides some good examples.
How Image Search is Changing Fashion Shopping
Tech CEO Ron Palmeri says the smart phone camera is on its way to becoming “the keyboard” because it can capture so much more information than words alone. Palmeri said he imagines a time when a shopper takes a picture of a desired item and rather than typing in a keyword to find it, they upload the image to a search engine that spits back a number of similar items at a range of price points or even items customized to a specific price point, if the user has integrated financial data into the model.
Palmeri’s firm, Layer, supplies the search technology for Nordstrom’s Trunk Club. The Twitter box describes the service and a posting shows one offer. The postings tend to show accessorized outfits, but a customized tux was the most recent post when I visited. Note that a personal fashion advisor is available on the mobile apps. Trunk Club actually started in 2009 as a service for men who didn’t like to shop but its expansion to women’s fashion and the importance of the smart phone camera is putting the emphasis on mobile.
 |
| Trunk Club Twitter Offer |
 |
| Trunk Club Twitter Page |
It’s about more than just selling; the importance of the images is influencing design at Nordstrom. “We are trying to understand how one pair of jeans plays out against another pair that was released in another season,” explained Justin Hughes, vp of product development and design at Trunk Club. “We want to get really granular and understand what really works.”
In a related development, Nordstrom is rolling out a line of stores with no inventory. The first store in Los Angeles has no merchandise, but it offers a fascinating array of services. See this post for a picture gallery that would stimulate a lively discussion about the future of retailing in the digital age.
British fashion supplier Apsos finds 80% of its UK traffic coming from mobile and 70% of its orders. Customers average 80 minutes each month on their mobile apps. No wonder Apsos is adding image search to its mobile apps! A customer takes a picture of an item or pulls one in from, say, Instagram for the search. The app returns similar items with reasonable accuracy, according to one reviewer. The results that don’t match so well may actually give customers more shopping ideas she says. There is plenty of room for ideation; Apsos currently has about 85,000 items in its image database.
 |
| Play the demo https://vimeo.com/226972601 |
Target has taken another route. The retailer is adding Pinterest’s Lens search technology discussed in the Social Commerce presentation) to its mobile app. Users take a picture with their smart phone and the search will find similar items for sale at Target. “The Pinterest partnership quite literally helps us shorten the distance from when our guests have an idea to when they’re ready to make a purchase,” said Rick Gomez, Chief Marketing Officer for Target. He adds that this technology will help understand what customers are looking for and therefore improve their merchandise planning.
What is the Future of Image Search?
There are a number of implications to be drawn from this discussion. As far shopping is concerned, many other visually-oriented categories like food should quickly be active in the space. The technology will continue to grow more powerful because of the players discussed here and other powerful ones, IBM for example, whose activities in deep learning were not discussed.
Brands have choices in terms of how to become a player in image-based shopping. Some will make a quick entry by partnering with a technology firm—Target and Pinterest, for example. Others will carefully build their own back-end processes, Apsos and eBay for example. Burberry is said to have engaged in massive image creation for a trial on Pinterest when it introduced its new Cat Lashes mascara. These activities are already beginning to impact the physical retail space as Nordstrom's new store illustrates.
Finally, search marketers will be pardoned if they emit another long sigh. Image search looks to be another disruptive development in the search space.
Look for all sorts of fascinating developments in the months and years to come!
Related Links
AI that trains itself
Asos new image search tool
Google ups its visual game
Retailers are experimenting with visual search
Visual search a top tool in 2019
