Meta has introduced an upgraded version of its ads recommendation system. The platform now uses advanced AI techniques to deliver ads to users. 

The new system focuses on understanding users' behaviors and interests by analyzing the order and context of their actions, such as clicks, likes, or purchases. Meta claims this upgrade will help advertisers reach their target audience more effectively and enhance ad relevance.

Meta’s previous ads recommendation system

Previously, Meta's ads recommendation system relied on Deep Learning Recommendation Models (DLRMs). This model used human-defined signals like clicks and page visits to understand user preferences. 

However, the older system had limitations, including missing details on the sequence of actions. For example, it couldn't differentiate whether a user clicked on an ad before or after visiting a page. It also couldn't fully capture how long users engaged with ads and relied heavily on manual input from humans to decide what data was important.

Meta’s new ads recommendation system

Meta’s new system uses sequence learning to recommend ads to users. The new system tracks event-based features (EBFs) to understand user behavior and interactions. 

Instead of relying on manually chosen data, EBFs automatically capture every detail of how users interact with ads. The system looks at each specific action users take, like clicking on an ad or visiting a page. It tracks details like the type of ad and when it was clicked, how long they engaged with it, and the sequence in which actions occurred. 

Sequence Learning looks at the order of users' actions to predict what kind of ads users might like next. For instance, if a user watches a video ad, and then visits a product page, the system can determine the relationship between those actions. When it tracks these interactions over time, the system can better predict the user’s future preferences.

The system uses advanced AI, inspired by language models, to understand users' actions. This helps show ads they’re more likely to care about.

Meta's upgraded ads recommendation system also allows the AI to track longer sequences of events. It captures changes in user preferences over time. Meta says the system is designed to get smarter as it processes more detailed data, with future improvements including better analysis of video and text content for even more recommendations.