In the world of online advertising, ad click prediction is a crucial task that helps advertisers reach their target audience effectively. However, overexposure to ads can lead to user fatigue and decreased clicks. To address this issue, researchers proposed "soft frequency capping," which limits the number of times an ad is displayed to a user based on their past behavior. This technique has been shown to improve ad click prediction in Yahoo Gemini Native, an online advertising platform.
The article explains that traditional frequency capping limits the number of ad displays for a specific user based solely on time-based intervals. In contrast, soft frequency caping considers the user’s past behavior and adjusts the ad display limit accordingly. This approach takes into account the user’s fatigue level and ensures that they see relevant ads at an optimal frequency.
The authors describe a real-world scenario where they applied soft frequency capping to improve ad click prediction in Yahoo Gemini Native. They analyzed user behavior data from different devices, including desktops, mobiles, and tablets, to develop a personalized fatigue level for each user. By taking into account this fatigue level, the authors were able to reduce ad fatigue and increase clicks by 20%.
To explain their findings more clearly, the authors use an analogy: "Imagine you’re at a cocktail party and keep getting served drinks by the same bartender. At first, it’s nice, but after a while, you start feeling like you’re being pestered. That’s similar to what happens when users see too many ads." By implementing soft frequency capping, advertisers can avoid overwhelming users and ensure they see relevant ads at the right time.
In summary, soft frequency capping is a valuable technique for improving ad click prediction by taking into account user fatigue levels. By personalizing the display limit based on each user’s behavior, this approach helps advertisers deliver targeted ads that are less likely to be ignored.
Computer Science, Information Retrieval