We don’t have to go in and validate that results are good, our customers are telling us the results are good.”
Global Search Lead
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.
Semantic Vector Search addresses low-performing queries without requiring curation of lexical rules. This is a major breakthrough that allows merchandisers to focus on more strategic initiatives. It’s a shift from reactive rule curation to smarter search, and proactive merchandising.
The traditional way of thinking about search recall and precision is based on a lexical concept of relevance – delivering explicit product matches to the terms entered. However, perfect precision often leaves dollars on the table. A semantic approach considers relevance from the perspective of the searcher’s goal, retrieving additional products to increase the value of the results to that shopper.
Two products that are semantically similar may have dissimilar performance. Why does product A sell more often than product B? By comparing the KPIs of semantically similar products, merchandisers can dig into the data to gain insight into varying performance and make informed decisions. Perhaps product B is consistently shown below the fold, or product B is priced too high.
Type-ahead suggestions are traditionally based on matching partial or complete query terms to popular searches, categories, and products. Use Semantic Vector Search to locate high performing queries and products that are nearby in the shared vector space. This powers a higher-converting experience that goes beyond understanding intent.
Semantic Vector Search can be combined with traditional filtering to find semantically similar products with specific features, such as a specific price range or high popularity and low return rate. Use this to power better recommendations and increase engagement.
Las Vegas travel search engine Vegas.com increased page views and engagement by 63% . Bounce rates for the mobile site dropped 8%. Conversion rates from search to a reservation or ticket increased by 33%. The team has since built the desktop search experience with Fusion as well. Full case study…
The American sportswear and footwear retailer put Lucidworks to work and saw a 10% lift in add-to-cart, improved personalization, and better analytics to drive merchandising decisions for promotional events.