2017-Search Ranking and Personalization at Airbnb

Table of Contents

1. Problem Definition
2. Methods
3. TODO & Questions & Further Reading

Problem Definition

1. Problem definition: Airbnb
   1. two-sided marketplace
   2. user never book the same listing for twice
   3. a listing can only be booked by one guest at the same time
   4. tradeoff of guest and host preference
2. Principle offline metrics: NDCG

Methods

1. Search components
   1. Retrieval of listing: location queries and broad queries
   2. Ranking
2. About user sessions
   1. label propagation backwords or forwards
3. Ranking model
   1. Learning To Rank model
   2. Utility
      1. booked 1, rejected -0.4, impression, 0, clicked 0.01
4. Listing embedding
5. Query embedding -> used in type-ahead

TODO & Questions & Further Reading

1. Dual Discount Curve Pairwise to push bad result bottom

• Previous
  2018-Applying Deep Learning To Airbnb Search
• Next
  2020-Improving Deep Learning For Airbnb Search