Heterogeneous graph neural network using offset temporal learning for search personalization

This invention describes a way to train an artificial intelligence system, specifically a heterogeneous graph neural network, to better understand users and the items they interact with. It works by looking at user-item interactions over two different time periods. The system is trained using data from an earlier period, but only for those user-item pairs that also show up in a later period. Once trained, this AI can then predict how relevant an item might be to a specific user.