Hierarchical attention time-series (hat) model for behavior prediction

A system, comprising:

• a processor;
• and a memory storing instructions that, when executed by the processor, cause the system to perform a method, the method comprising:
• receiving, by an application, activity data of a user;
• identifying, by the application, user sessions comprising sets of time-stamped actions in the activity data;
• segmenting, by the application, the activity data into subsets corresponding to the user sessions;
• providing, by the application, the subsets as inputs to a hierarchical attention time-series (HAT) model that comprises:
• a first layer that determines attention scores for respective time-stamped actions in the subsets;
• and
• a second layer that determines attention scores for the subsets based on aggregations of the attention scores for the respective time-stamped actions;
• and
• receiving, by the application, as outputs from the HAT model in response to the inputs:
• a prediction based on the subsets, the attention scores for the respective time-stamped actions, and the attention scores for the subsets;
• and
• explanatory information based on the attention scores for the respective time-stamped actions and the attention scores for the subsets.

A method, comprising:

• receiving activity data of a plurality of users;
• identifying user sessions comprising sets of time-stamped actions in the activity data;
• segmenting the activity data into subsets corresponding to the user sessions;
• associating the subsets with labels based on whether given users of the plurality of users performed a given action;
• and
• training a hierarchical attention time-series (HAT) model using the subsets and the labels, wherein the HAT model comprises:
• a first layer that determines attention scores for respective time-stamped actions in the subsets;
• a second layer that determines attention scores for the subsets based on aggregations of the attention scores for the respective time-stamped actions;
• and
• an output layer configured to output:
• a prediction corresponding to the given action that is based on respective actions that are provided as inputs to the HAT model and outputs from the first layer and the second layer;
• and
• explanatory information based on the outputs from the first layer and the second layer.