LF1.io
Published in

LF1.io

Introducing the AI driven product fingerprint of Aleph Search

At Aleph Search we are often asked by new customers: “how are you able to simultaneously offer text-search, image-search, recommendation as well as additional applications?” The answer, which lies at the heart of our technology, is our proprietary Product Fingerprint algorithm. In developing this algorithm we have drawn on several current strands in artificial intelligence research:

Aleph Search’s product fingerprint lies at the intersection of several technologies

Our product fingerprint algorithm allows us to excel where traditional technologies falter. Traditional technologies entail the following:

  • Text-search means matching query texts to product attributes
  • Image-search means taking one image and finding individual images which match on certain products
  • Product recommendation means finding products with similar attributes

In addition, no universal framework is available to facilitate a range of additional work flows on e-commerce platforms.

In contrast to these traditional technologies, our product fingerprinting allows us to:

  • Deliver text-search so that users can search directly through the content of products, as reflected in the images, texts and attributes on products
  • Deliver reverse image search which understands the content of the query image and matches this with the content of the product (as reflected in the images, texts and attributes)
  • Deliver similar product recommendations where the notion of similarity fits the commerce domain: similar fashion items look similar, similar electronics have similar specs., similar books are about similar topics etc..
  • Deliver multiple additional bespoke AI work flows specific to the domain in question.

What is a product fingerprint?

A product fingerprint is a unique numerical identifier of a product which encapsulates the value proposition presented by a product to a consumer.

As an example, look at the fashion item in the image below. Users will judge this item based on the brand, the look, the perceived fit, the description of the item, its attributes and the list of materials — everything visible on the page. The product fingerprint of this item implicitly summarizes and condenses this information.

A product’s appeal lies in the presence of multiple pieces of information

How does product fingerprinting work?

The product fingerprint algorithm is built based on our own internal stack which includes elements of deep learning, NLP, computer vision and information retrieval technologies. It’s a technology which has been in development for a long while — however to summarize we can take a look at the figure below.

Aleph Search’s product fingerprints arise from rasters of deep neural networks

Our technologies includes rasters of deep neural networks (these are the nets visible on the image) which perceive and combine the content from each element of a product’s data. This combination comes out as a long list of numbers (visible on the right hand side) — this list is the product fingerprint. Because the list isn’t too long — a huge amount of compression occurs in the deep neural networks — it can be manipulated very quickly and adeptly in various applications. So it plays a central role in text-search, image-search, recommendation as well as a host of additional AI products we’re developing.

Outlook

We’re working on applying product fingerprinting to products in fashion, furniture, electronics and more. Additional applications are presenting themselves on a weekly basis, so watch this space.

--

--

Charting the way forward for AI

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store