Ads Sensitivity Modeling in Instagram Ads

Taiyuan Zhang

Keywords: causal inference, causal learning, online advertisement, user sensitivity

TL;DR: How we made Ads Sensitivity Modeling work in Instagram Ads

Abstract:

In the realm of social media advertising, a del-
icate balance must be struck between ads value,
and user engagement. To achieve an optimal equi-
librium, sophisticated ad supply strategies are re-
quired, which involve deciding the best time, place
and density to show ads to individual users. Pre-
dicting users’ sensitivity towards ads has been a
critical component of the strategy.
This paper presents a case study on how Instagram
Ads harnesses the power of causal inference ma-
chine learning, to learn each user’s sensitivity to-
wards ads, which is then leveraged to enable ads
supply personalization. This is an extremely chal-
lenging task due to massive scales and subtlety of
causal effects from supply changes. By leverag-
ing data from RCTs and SOTA causal inference
techniques, we demonstrated that our framework
enables accurate estimation of user ads sensitivity
over extended periods, through rigorous offline and
online studies.

Submission Number: 5