• by Laurie Sullivan  @lauriesullivan, May 5, 2023

Bloomreach launched an algorithm for its Discovery product, intended to increase search relevance for B2B companies.

The platform -- developed by the commerce experience cloud company -- uses natural-language processing (NLP), an algorithm that understands numeric product attributes in a search query, such as "8x10" frame or "3/4-inch wire." The algorithm can return results with precise matches.

It supports English and non-English languages, so it can account for different measurement systems and ways of writing numeric values.

Raj De Datta, CEO and co-founder of Bloomreach, in an earlier interview said generative AI use cases might be for website content or outbound outreach like emails. “We also use different AI models for language translation,” he said. “In our search engine you can search in multiple languages. The system can learn the semantics and attributes in multiple languages.”

The search algorithm from Bloomreach Discovery is based on the company's industry-leading semantic understanding, which leverages NLP to differentiate product types from product attributes.

Other common use cases include -- but are not limited to -- using quotation marks for inches, comma in queries, and superscript in queries.

Most traditional keyword-based search engines cannot decipher these types of numerical attributes in a search, meaning that search owners have to manually enter synonyms and manage misspellings to account for the variations, per the company..

When asked about bias and challenges with LLMs, De Datta said the companies building and using the platforms need to be responsible. “There are a lot of dangerous things in the world,” he said. “We can shut off the APIs, but LLMs are not going away.”

Bloomreach doesn’t just take GPT technology and embed it in its products, he said. There are many guardrails and restraints to govern what is requested in queries.

“We might use customer profiles that we understand to create relevant results back for those customers,” he said, ensuring that data is built on a specific company’s information and is not shared.