Architecture of relevancy search at mobile.de
summary
mobile.de runs ~1.6 million active adverts and collects 300 million interactions per day into Hadoop, aggregates 30-day profiles in Cassandra, and serves a REST price-rating microservice (H2O) that labels listings. Search enriches queries with user preference features and uses Elasticsearch plus learning-to-rank with RankLib and the Elasticsearch LTR plugin to rescore roughly top 100 results by a click-trained model; outcome: improved median first-click position and higher dealer contact rates.
AI-written summary · about
chapters (9)
- 0:00Introduction to mobile.de
- 4:25Incremental approach to relevancy
- 5:28Tracking user preferences
- 7:12Relevancy search 1.0
- 8:36Price rating for fair offers
- 11:46Learning to rank for relevancy
- 16:04How learning to rank works
- 18:57Results and future outlook
- 20:02Q&A
related talks
27mA smooth ride: Online car buying and selling at mobile.de
25mBetter search relevance using Learning to Rank at mobile.de
39mBuilding a personalised home feed for Germany’s biggest classifieds website
43m“Which car fits my life?” - mobile.de’s approach to recommendations
40mLevel Up: Everything you need to know about search, ...
39mRanking Listings with Machine Learning and ElasticSearch
26mStreamlining Property Search Across Multiple Platforms
44mRobust AI Search Ranking for Radical C2C Marketplace Growth