The foundational approach to precision, recall, and relevance is churning out a lackluster ecommerce search experience: too many zero results, irrelevant results, results excluding many relevant products. Plus, it takes too many searches to fulfill the customer’s goal.
We’ve painted ourselves into a corner with ecommerce search experiences that are driven by a lexical, or literal, approach, that requires matches between keywords and the product index. If the search engine strays from precision we quickly wrangle it back in with business rules. Search is on a tight leash because of high customer expectations and poor experiences. Hint: tightening up the leash is not fulfilling your customer’s goals.
A recent study from Baymard found that search is so poor that 31% of searches ended in vain.
“And among the top 60 e-commerce sites, a whopping 70% of the search engines are unable to return relevant results for product type synonyms – requiring users to search using the exact same jargon as the site – while 34% of the sites don’t return useful results when users search for a model number or misspell just a single character in a product title.”
Fulfilling customer goals sounds easy, and it can be! But it’s going to take a different approach to precision, recall, and relevance. It’s also going to take a different approach to interpreting queries, from lexical to semantic vector. This evolution in product discovery is made possible by the recent launch of Lucidworks managed service, Never Null, which leverages a combination of mature semantic search strategies and dense vectors. The easy part here is that nothing is manual, this is a true AI-driven approach that gains intelligence from the exhaust of your shopper behavioral signals.
What Is Semantic Vector Search?
Here’s the technical explanation:
- Semantic vector search uses deep learning to associate products with queries in a shared semantic vector space. An encoder model learns from product discovery signals to encode products and queries as vectors. Incoming queries are encoded on the fly, and then products that are “near” the query in the shared vector space are returned.
The practical explanation:
- A vector space is a way of organizing products and queries based on their similarities to each other, relative to the rest, not just their lexical similarities. Think about a fish counter that has all the fresh seafood, that’s one relationship among the products. The seafood is organized by shellfish and fish meats, another relationship. Lastly, the lemons, cedar boards, and OldBay are at the counter, because they are peripherally related.
The visualization below abstracts this by showing concentrations or groups of colors representing the different categories of products in relation to each other.
Semantic vector search considers relevance from the perspective of the customer’s goal or intent, retrieving additional products to increase the value of the results to that shopper. Where lexical search falls short offering too many (or not enough) search results, semantic vector search delivers.
Slash Zero Results Queries
A common problem is that shoppers are often searching for products that you sell, but your current approach can’t connect the dots between the query and your products. Whether it’s a brand you don’t carry, out-of-stock products, or nomenclature, semantic vector search will return the most relevant products to fulfill that shoppers goal—no manual intervention or rules required. Real-time query interception unlocks a gold mine of value for your shoppers and your business, converting zero results into add-to-carts.
Notice in the above search that the “shop vac” query returned zero results. That doesn’t mean the store doesn’t have shop vacs, it means that nowhere are the products called “shop vac” and so lexical search falls short on returning any results. Without rules or any other manual intervention, semantic vector search can learn from your persistent shoppers and retrieve relevant products for previously zero result queries.
How To Balance Precision and Recall
This is a strategy that may require some consideration and negotiation internally at your company. Do you want to deliver exactly what the shopper is asking for? Or do you want to understand their intent and fulfill their goals? At the surface, this sounds like an easy answer but as mentioned before, it’s a fundamental change to today’s definition of precision, recall, and relevance.
Semantic vector search will enrich the results with additional related products to reduce friction in the product discovery journey, shorten the path to purchase, and show your shoppers you understand their goals. In this case, it’s making a delicious, well-seasoned piece of salmon.
Finally, a Solution for Zero Result Searches
Your customers have high expectations, limited patience, and money to spend. Semantic vector search eliminates false zero results and accelerates the path to purchase. With Never Null, now you can:
- Easily map similar products within your catalog so your ecommerce experience will rarely surface a zero results page—if at all!
- Stop chasing zero result queries and free your search managers and merchandisers from the pain of manual rule creation.
- Avoid replatforming until you’re ready. Connect the model to your existing product discovery platform and see results quickly.
You can also check out our on-demand webinar: Say Goodbye to Zero Results to learn how top retailers are using semantic search to solve the “null results” problem.