Combating Ad Fatigue via Frequency-Recency Features in Online Advertising Systems

Online advertising is a driving force of Internet services today. One of the main challenges in advertising systems is finding the right balance between user experience and overall revenue. In this paper we address one of the problems that negatively impacts the user experience, specifically, the repeated display of identical ads to the same user. This problem leads to the phenomenon called "ad fatigue", characterized by diminished interest in the ad, resulting in a decrease in the click-through rate (CTR) as users encounter the same ad repeatedly. Naïve solutions such as placing a hard limit on the number of times a specific ad is displayed to a specific user, usually come at the cost of reduced revenue. To address the ad fatigue problem, we introduce a new family of features, called FoRI (Frequency over Recent Intervals). FoRI features integrate information about frequency and recency of previous user-ad interactions within the CTR prediction model. This approach involves allocating these interactions to unevenly distributed time intervals, enabling a higher emphasis on more recent interactions. Furthermore, we introduce new metrics to assess ad fatigue in terms of the repetitiveness and novelty of the displayed ads. We conducted a comprehensive large-scale online evaluation which shows that integrating FoRI features into our CTR prediction model offers two-fold benefits. Firstly, it improves user experience by reducing the occurrence of repeated ads by 15%, and increasing the exposure to unseen ads by 5% (ads not previously displayed to the user), leading to a substantial boost in CTR. Secondly, it significantly increases revenue.