Optimal signals assignment for eBay View Item page

TL;DR

The paper tackles optimizing the assignment of textual and visual signals on eBay's VI page to boost business metrics despite extremely small uplifts and strong signal correlations. It presents two production-ready ranking approaches—Retrospective Learning (conditioning on conversion to predict shown signals) and Conversion Likelihood Estimator (ranking by per-qualification-set conversion rates)—plus a novel offline uplift metric based on Pearl's causal adjustment to enable model selection from randomized data. Online A/B tests show significant gains: Retrospective Learning increases Add to Cart by $+0.36\%$ and Best Offer by $+0.49\%$, while Conversion Likelihood Estimator improves GMB and Organic Revenue, indicating enhanced engagement across the purchase funnel. The work provides a practical evaluation framework and demonstrates the feasibility of deploying signal optimization on the VI page with a robust offline-to-online validation pipeline, offering actionable paths to further lift the VI-page impact.

Abstract

Signals are short textual or visual snippets displayed on the eBay View-Item (VI) page, providing additional, contextual information for users about the viewed item. The aim in displaying the signals is to facilitate intelligent purchase and to incentivise engagement. In this paper, we present two approaches for developing statistical models that optimally populate the VI page with signals. Both approaches were A/B tested, and yielded significant increase in business metrics.

Figures (1)

• Figure 1: Illustration of shown signals in randomization and prediction