Meta Platforms, Inc.
Determining a target device profile including an expected behavior for a target device

Abstract:

Techniques for obtaining a target device profile included an expected behavior for a target device are disclosed. An unsupervised learning algorithm is applied to a global dataset including device data corresponding to multiple device types. Additionally, the unsupervised learning algorithm is applied to a device type dataset including device data corresponding to a single device type. Clusters are obtained from both the global dataset and the device type dataset. Clusters that share a device attribute with a target device are identified as "relevant clusters." Behaviors associated with the relevant clusters are used to determine expected behaviors for the target device. Values, for a particular behavior that is common to multiple relevant clusters, are merged to determine an expected value, for the particular behavior, for the target device. Additionally or alternatively, the behaviors, associated with the relevant clusters, are aggregated to form the target device profile.