Optimizing Retail Search Using Natural Language Processing
One of the largest home improvement retailers had a messy online search which created customer confusion slowing sales.
Each month this company added hundreds of new products to their large inventory of over 2 million unique SKUs. It is not feasible to tag every product and anticipate every customer keyword search combination. Products searched often fall into multiple categories and, depending on the instance, the term may have multiple meanings. For example, “Door” can be a standalone door or a feature of another product, and searching “coffee table” would return a coffee maker, table saw, and brown tables (coffee colored).
The Lymba Solution
Process
To optimize retail search, we created a company-specific ontology with in-house terminology. We were then able to leverage that knowledge to recategorize the backlog of products by department. ur system was able to better understand what customers were looking for, and provide a better understanding of the offered products.
Outcome
This company improved their product tagging from 68% to 93% (as measured by client SME). This kept customers happy and enabled them to find the goods they were looking for easier.