Paper Review — Ranking at Scale at Booking.com

Paywall

Full text unavailable — this is a Medium member-only post. Summary based on publicly visible content only. Original article: https://medium.com/@joparga3/paper-review-ranking-at-scale-at-booking-com-47978fa6d08d

Read the primary source instead

This post reviews an open-access paper. The full text of that paper is freely available, and the vault covers it in depth at Beyond Algorithms - Ranking at Scale at Booking.com.

Jose Parreño, a Sr. Data Science Manager at Skyscanner, reviews the 2020 Booking.com paper Beyond algorithms: Ranking at scale at Booking.com. He describes it as “a gold mine of how to start with any ranking project,” arguing its value is not limited to organisations operating at Booking.com’s scale.

Scope of the review

The visible portion establishes that the review covers two of the paper’s dimensions:

Modelling

  • Defining target variables
  • Feature engineering
  • Biases in ranking
  • Offline model evaluation

Experimentation

  • Leakage

The one substantive visible claim

Parreño’s framing point is that defining a target variable is difficult in ranking and recommendation, and that this is the foundational obstacle rather than a preliminary detail — “Without a target variable, you simply cannot run build a ML model.” The body cuts off at that sentence, before the explanation of why it is difficult.

This matches the emphasis of the underlying paper, whose signal-definition section weighs candidate labels on satisfaction, volume, delay, and bias — see Ranking Signal Selection.

People