How to select the best set of ads: Can we do better than Greedy Algorithm?

TL;DR

This paper introduces the Greedy-Power Algorithm, an enhancement over the standard greedy approach, to select the optimal set of ads by iteratively improving solutions and avoiding local optima, leading to better ad selection outcomes.

Contribution

The paper proposes a novel Greedy-Power Algorithm that iteratively refines ad set selection, outperforming the traditional greedy method in maximizing clicks.

Findings

Greedy-Power consistently outperforms the standard greedy algorithm.

The algorithm guarantees non-worse solutions, only improving or maintaining the goal function.

Simulations show it escapes local optima more effectively.

Abstract

Selecting the best set of ads is critical for advertisers for a given set of keywords, which involves the composition of ads from millions of candidates. While click through rates (CTRs) are important, there could be high correlation among different ads, therefore the set of ads with top CTRs does not necessarily maximize the number of clicks. Greedy algorithm has been a standard and straightforward way to find out a decent enough solution, however, it is not guaranteed to be the global optimum. In fact, it proves not to be the global optimum more than 70% of the time across all our simulations, implying that it's very likely to be trapped at a local optimum. In this paper, we propose a Greedy-Power Algorithm to find out the best set of creatives, that is starting with the solution from the conventional Greedy Algorithm, one can perform another Greedy Algorithm search on top of it, with the option of a few or even infinite rounds. The Greedy-Power algorithm is guaranteed to be not worse, as it only moves in the direction to increase the goal function. We show that Greedy-Power Algorithm's performance is consistently better, and reach the conclusion that it is able to perform better than the Greedy Algorithm systematically.