Microsoft is working on Decision Service, which uses machine learning to process user clicks and views in real time. The end goal is to provide a platform that is capable of analyzing the click and viewing activity of the user in real time, to deliver a personalized and optimized ad experience in real time. The platform makes the decisions to optimize the user experience by using metric markers that are programmed into the system.
At the Bing Ads Next event in Redmond, Washington, Gal Oshri, Microsoft Research program manager, and LiHong Li, Microsoft Research senior researcher, explained how the new platform works. Testing with MSN, about 20 articles were chosen by the editorial team. The articles ranking showed how the clicks and views on the articles worked its way through the platform in near real time, and produced a customized model every 5 minutes. This led to an increase of almost 27% in the click through rate of the articles, markedly improved the overall engagement and interaction with the page, which indirectly affected the overall advertising clicks by 9%.
The research on the above model seems to resolve a lot of the issues that real time traditional machine learning typically has difficulties with. Gal Oshri states “This is important for applications like News, because when a new story gets published you need to react quickly. A breaking news story can become less popular in less than an hour.”
Another way that Microsoft has tested the platform is to not program the first ad that the user sees. Instead, after the first ad is clicked, it is after the second click that the real-time optimization occurs. The second click is when the platform makes the choice of what to display.
Privacy is also a key consideration. Each publisher or advertiser gets a customized version of the platform for use with their application. The user’s personal data is not stored by the platform, rather, it is stored by the application, keeping user data private.