Mihajlo Grbovic

ML/Research Scientist at Airbnb. Led two major search ranking initiatives: listing embeddings from 800M+ click sessions, and the ML-powered Experiences ranking system that delivered ~28% cumulative booking improvement.

Articles

• Listing Embeddings in Search Ranking — 32-dim listing embeddings from 800M sessions, EmbClickSim/EmbSkipSim features, +21% CTR
• Machine Learning-Powered Search Ranking of Airbnb Experiences — four-stage ML progression (GBDT → personalization → online scoring → business rules)

Key Contributions

• Adapted word2vec to listing click sessions; booked listings as global context across sessions
• Market-specific negative sampling to address geographic concentration bias
• Real-time personalization features EmbClickSim and EmbSkipSim in main LTR model
• Business rules via training data weighting (no separate rules engine)

Related

• Airbnb · Learning to Rank · Personalization · Embedding Fine-tuning · Dense Vector Retrieval