We are pleased to announce the general release of Constructor’s Cognitive Embeddings Search. Drastically reduce your zero-results rate by expanding results and increasing your search coverage!
You are in the market for a small table to set your phone and water bottle near your bed.
So, you pop open your laptop and go to your favorite furniture retailer’s website.
Inundated with a myriad of fun and tasteful furniture assortments on the home page, you decide to cut the chase short by using the search bar. You type in “bedside table” and hit enter.
What? That can’t be. Surely a furniture store carries those kinds of tables.
You hover your mouse over the navigation bar, studying and comprehending all the different categories of furniture. You go to “Bedroom” first, then scan down the options. Mattresses… Headboards… Wardrobes… Dressers…
Finally, you find the “Nightstands” tab. But by this time, you’ve gotten so tired and frustrated that you close the browser and go back to sleep.
The next day you search for “bedside table” on Amazon.com. There, you get over 7,000 results, instead of 0 and order your new nightstand immediately.
The situation above is a classic case of the problem with zero-results queries. When a search provides no results, the risk of site abandonment sky-rockets.
Many eCommerce retailers spend hours upon hours each week setting up synonyms and redirects manually to reduce the number of zero result searches, often leading to multiple internal teams working on the same issue. However, in this ever-changing world where new search phrases and trends pop up by the second, how much time can realistically be dedicated to this labor-intensive procedure?
Constructor can now dramatically reduce the zero-results rate for our eCommerce partners and provide a comprehensive and smoothly designed product discovery experience through the native addition of our cutting edge Cognitive Embeddings Search.
At its core, Cognitive Embeddings Search features a vector-based search algorithm and draws upon layers of deep learning to transform a product catalog representationally into a “sea of stars”. At a high level, if we think about each item in a catalog as a “star,” we can then visualize a product catalog as a “sea of stars” formed by many, many items. From this we can understand the “stars” better by clustering them and extrapolate which ones are the closest neighbors to a given search query through this representation.
For example, we can understand the relationship between the search term “healthy snacks” and the result “no sugar added freeze-dried bananas” by training our Cognitive Embeddings model on product data, such as categories, product name, and text descriptions. Therefore, the algorithm maps out the relations from “healthy snacks” to “healthy foods” to “healthy & organic dried banana chips” to finally “no sugar added freeze-dried bananas” and narrows down the search engine’s understanding of “healthy” by mapping a specific term to neighboring concepts. And inversely, if “no sugar added freeze-dried bananas” are not available in a product catalog, we can return other healthy snacks, like “organic dried banana chips” instead.
Since there is always a closest “star” to a “point in space”, we are able to serve results for every search query, reducing the null results rate to virtually 0%.
Cognitive Embeddings Search improves search performance by greatly reducing zero-results rates with contextual awareness through vector/spatial representation of a catalog.
With these added capabilities from Cognitive Embeddings, we can serve even more attractive results when, for example, a user searches for “healthy peach snack”, and we return “no sugar added” and “organic” snack options while also disregarding the word “peach” should there be no peach items in the product catalog.
Combining cognitive embeddings’ contextual awareness with Constructor.io’s KPI-lifting search creates an unparalleled eCommerce product discovery experience.
If the furniture site in our earlier example had Constructor’s product discovery with Cognitive Embeddings Search enabled, you would have quickly found a “bedside table”. This allows our hypothetical furniture retailer to reduce reformulated and frustrated searches, increase revenue from additional product discoverability, and enable a well-rounded user experience.
From a more technical perspective, not only is Cognitive Embeddings Search stellar in zeroing the zero-results rate and helping users discover more products, it is extremely performant. The average inference time (total response time from when the search query is received to the calculations made on the backend to the results being served to the end user) is ~3ms across our diverse customer base.
Additionally, Cognitive Embeddings Search specifically solves for these types of search queries and search intents:
By combining the existing behavioral data approach to search with extremely performant contextual awareness that is especially helpful for long-tail queries with little behavioral data makes Constructor search a powerhouse in the eCommerce product discovery field.
From our initial rounds of A/B testing, we have some exciting results to share!
Please contact your customer success manager if you are already a Constructor partner and are interested in learning more about Cognitive Embeddings Search. Otherwise, please click here to request a demo to learn more about Constructor’s search and product discovery solutions.