Ecommerce ops · Production

Whatnot enhances search with GPT-powered query expansion and spell correction

The problem

Misspelled search queries like 'jewlery' produced nearly empty results pages, causing users to falsely conclude Whatnot lacked relevant content. Acronym and abbreviation queries such as 'lv' or 'nyfw' also returned low result counts and lower downstream engagement.

Workflow diagram · grounded in source

Search query data collection

trigger

“We begin by collecting search queries from logging, such as "funko pop," "fine jewelry," and "nyfw."”

Tokenization and frequency filtering

validation

“We gather frequently occurring tokens by summarizing their usage over the past 14 days. Specifically, if a token has been utilized in search queries more than 3 times during this period, we consider it as a token to be included in the su…”

GPT spell correction and expansion

ai_action

“we send them to the GPT model along with a prompt designed to identify potential misspellings and to suggest expansion text from acronyms/abbreviations. Since the model is trained on such a wide, large collection of data, it has knowledg…”

Cache population with corrections

integration

“we put them into our query expansion cache. This is a tier in a production-level key-value store that maps from original query tokens to the lists of potential corrections/expansions, along with their associated confidence levels.”

Real-time query expansion lookup

integration

“we refer to the query expansion cache to identify potential spelling corrections and abbreviation expansions related to the tokens of the user's query”

Augmented search result page

output

“we generate a search result page from the combination of the original user query and the expanded queries retrieved and processed from our cache based on their confidence levels”

Reported outcome

The GPT-based query expansion approach reduced irrelevant content by more than 50% for misspelling and abbreviation queries compared to the previous method, while also yielding substantial improvements in query expansion accuracy and streamlining the generation and serving process.

Reported metrics

Irrelevant content reduction for misspelling/abbreviation queriesmore than 50%

Query expansion accuracysubstantial improvements

Generation and serving process efficiencystreamlining the generation and serving process significantly

Reported stack

GPT

Source

https://medium.com/whatnot-engineering/enhancing-search-using-large-language-models-f9dcb988bdb9

Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

The GPT-based query expansion approach reduced irrelevant content by more than 50% for misspelling and abbreviation queries compared to the previous method, while also yielding substantial improvements in query expans…

What tools did this team use?

GPT.

What results were reported?

Irrelevant content reduction for misspelling/abbreviation queries: more than 50%; Query expansion accuracy: substantial improvements; Generation and serving process efficiency: streamlining the generation and serving process significantly (source-reported, not independently verified).

How is this ecommerce ops AI workflow structured?

Search query data collection → Tokenization and frequency filtering → GPT spell correction and expansion → Cache population with corrections → Real-time query expansion lookup → Augmented search result page.