Performance-efficient system for predicting user activities based on time-related features

摘要

A recommender system uses an activity decision tree to model the changes in a user's behavior according to a plurality of time-related features. The system determines historical activities for the user, and generates a decision tree for the user's historical activities. Each leaf node of the decision tree is associated with an activity-prediction model that computes a probability for a corresponding activity. The system selects a path of the decision tree from a root node to a leaf node of the decision tree based on a target time. The selected path traverses two or more non-leaf nodes that are each associated with a temporal decision model that compares the target time against a temporal classifier. The system then determines a probability for a user activity based on an activity-prediction model of the selected path.