2023- Learning To Rank Diversely

Table of Contents

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

Problem Definition

1. Prove the following assumption is wrong: the booking probability of a listing could be determined independently of other listings in search results.
2. Key metrics: NDCG
3. About affordabilityleaning and quality: we can apply the broad classification that the majority ∼80% of users are affordabilityleaning. The remaining minority ∼20% are quality-leaning

Methods

1. Previous methods
   1. check section 6
2. Current methods
   1. model for each position considering previous listings

Impact

1. NDCG
   1. Offline: +0.45%
   2. Online: +1.5%
2. Booking & booking value: 0.29% in booking and 0.8% in booking value (it means it mainly influence high value booking, right??)
   1. mainly come from first time user
3. Engagement: 0.46% list views; 1.1% in list saved
4. Position utilization
   1. the utilization of the top position increased significantly, followed by some gain for the second position

TODO & Questions & Further Reading

1. paper: Managing Diversity in Airbnb Search
2. re-visit introduction section -> summary needed

• Previous
  2019-Neural Product Retrieval at Walmart.com
• Next
  2018 - Customized Regression Model for Airbnb Dynamic Pricing