Optimizing Query Evaluations using Reinforcement Learning for Web Search

TL;DR

This paper introduces a reinforcement learning approach to optimize match planning in web search, reducing index access by up to 20% without significantly impacting candidate quality.

Contribution

It formulates match planning as a reinforcement learning problem, achieving more efficient index scanning in web search systems.

Findings

Up to 20% reduction in index blocks accessed.

Minimal or no degradation in candidate set quality.

Demonstrates effectiveness of RL in search engine optimization.

Abstract

In web search, typically a candidate generation step selects a small set of documents---from collections containing as many as billions of web pages---that are subsequently ranked and pruned before being presented to the user. In Bing, the candidate generation involves scanning the index using statically designed match plans that prescribe sequences of different match criteria and stopping conditions. In this work, we pose match planning as a reinforcement learning task and observe up to 20% reduction in index blocks accessed, with small or no degradation in the quality of the candidate sets.