Wallapop
Classified-ads marketplace in southern Europe (Spain, Italy, Portugal), with more than 70 million users. Almost anyone can sell almost anything, so its search runs over user-generated listings with frequent misspellings, and it also has a locality filter so buyers can find items near them. The search team includes search engineers, a principal search engineer, a product manager, front-end engineers, data scientists and analysts.
Search work
- Zero Results mitigation: the team had already contributed a Spanish stemmer to Lucene and shipped synonyms and hyponyms, Spelling Correction, a wider geographic radius for some queries, and looser Boolean matching (minimum-should-match on Solr). Its next step was alternative queries through Query Relaxation. The first version drops the least-frequent query term. It reached about 80% offline accuracy, and an A/B test was about to start as of October 2023.
- Vector search was planned as a medium-term investment. Embedding a large, fast-churning listing inventory was the main cost concern.
Videos
People
- Jean Silva — search engineer