We asked Nilay Oza, Co-Founder & CEO of partner Klevu, to take us through the application of NLP in their site search solution for Magento Commerce.
Natural Language Processing (NLP) is a subfield of Artificial Intelligence and focuses on the interactions between humans and computers. NLP looks at how to program computers to process and interpret large amounts of data around human language.
If you’re an ecommerce manager, a solutions specialist, a product manager or an AI retail tech enthusiast, you’ll love this mini series of articles! They aim to demonstrate how NLP is applied by Klevu to provide relevant site search experience for online stores. NLP powers Klevu at multiple levels. In the previous article we explained Level 1: Catalog Processing, in this article we focus on Level 2: Query Processing.
Level 2: Query Processing
In Level 1, Catalog Processing, Klevu prepares the catalog for search. But there’s more work to be done. As a second step, Klevu needs to understand the terms of the query in real time – an extremely important step.
The application of NLP in Klevu allows for Smart Query Processing. Query processing involves understanding the keywords shoppers use to search for specific and relevant products within the search bar. The main components of query processing include identifying the subject from the query, normalisation of specific terms, handling errors in the query, and extracting relevant products for the shopper. Klevu’s query processing happens instantly, in real time, and identifies the intent of the shopper instantaneously.
Query processing finishes shoppers’ thoughts on what they intend to discover in the store.
Query processing finishes the thoughts of the shopper by visually showing relevant suggestions, categories and products based on keywords. Even after the first letter, Klevu will make suggestions for the shopper to choose from, making it really easy for customers to search and find the products they want. Examples below shine further light on the importance of query processing.
Search terms such as ‘formal tops under 100 USD’ requires query processing in real time to parse on the intent of the shopper. In this example, the shopper not only wants to find formal tops, but also wants to find ones that are priced under 100 USD.
The figure below shows an example from Klevu client Made.com, where a shopper looks for ‘grey sofabed’ but with specific price constraints (‘under 800’).
Note how Klevu query processing identifies the specific colour and still filters results by pricing.
Shoppers search in several ways. At times, the search term itself is quite short but requires processing to bring relevant suggestions. Consider the following example:
A shopper starts to type ‘bed sheets’ but before they finish, Klevu powers a suggestion for ‘beds’ and ‘bed sheets’ once they’ve typed only ‘bed’ or even ‘be…’.
The figure below shows an example from Klevu client Bathroom Takeaway. The shopper’s just entered the word ‘shower’; Klevu not only produces relevant suggestions but through query processing, it shows relevant categories (bath shower screens, bath suits, mixer taps) and keywords that would finish the thought process for the shopper.
Sometimes, two nouns are used by shoppers in search but without a space, for example, ‘fairylights’. In this case, by processing the query, Klevu would first decompound the query to identify relevant results from the catalog. The second intent parsing would allow Klevu to learn that the shopper is after ‘lights’ as a primary noun. Understanding that the search term has two nouns, and the fact that the second noun is the primary noun, is all established through linguistics.
Shoppers often want to find products with good reviews or ratings from previous shoppers (proof the product is worth the cost) and might search for something like ‘top rated bags on sale’. Klevu will recognise what they are looking for and present them with relevant top rated products only. Klevu in this case, will look into ratings supplied in the catalog and determine high ratings as popular products. ‘On sale’ will also be incorporated into the query to find products that have a sale price lower than the original price.
To summarise query processing, the figure below demonstrates how Klevu understands and breaks down a search query to bring relevant top-rated results to shoppers.
This series of articles provides an insight into NLP (Natural Language Processing) capabilities of Klevu at two levels: catalog processing and query processing. The articles provide real world examples to shed some light on how processing of catalog and query helps to bring relevant results to shoppers. Klevu provides one of the most comprehensive, innovative applications of NLP into the retail domain. The ultimate aim of Klevu is to form a dialogue with shoppers and bring discovery to a new level, where they can find products through natural, open-ended interactions. Machine Learning, combined with NLP is at the core of Klevu to ensure that retailers stay cutting-edge when it comes to understanding the intent of the shoppers when searching their store. In this journey, we thank our partners The Pixel (www.thepixel.com) for providing us with a platform and an opportunity to offer these insights into the world of NLP-based search. To learn more about the application of NLP in retail, join us at one of our free online training sessions, registration available at Klevu.com. Happy shopping!