Account behavior prediction using prediction network

A non-transitory computer-readable storage medium containing instructions, that when executed, control a computer system to be operable for:

• receiving, by a computing device, a value for a prediction network that is based on a number of consecutive playbacks of instances of content from a viewing behavior;
• inputting, by the computing device, a sequence of historical behaviors for a plurality of instances of content into the prediction network to generate a sequence of values that model the sequence of historical behaviors, wherein a number of adjacent nodes in the prediction network are analyzed in an operation, and wherein the number of adjacent nodes is set based on the value for the number of consecutive playbacks of instances of content from the viewing behavior;
• generating, by the computing device, a sequence of attention scores based on a similarity of a current behavior for a first instance of content to respective instances of historical behaviors in the sequence of historical behaviors;
• adjusting, by the computing device, respective values in the sequence of values based on corresponding attention scores in the sequence of attention scores to generate an adjusted sequence of values;
• and
• sampling, by the computing device, the adjusted sequence of values to generate an output from the prediction network that models the sequence of historical behaviors based on the current behavior, the output for determining a prediction if the current behavior is indicative of the viewing behavior of the number of consecutive playbacks.

A method comprising:

• receiving, by a computing device, a value for the first prediction network that is based on a number of consecutive playbacks of instances of content from a viewing behavior;
• generating, by the computing device, a first output using a first prediction network, wherein a first input to the first prediction network is based on a first number of behaviors within a first time threshold, and wherein the first prediction network analyzes a number of adjacent nodes in an operation, and wherein the number of adjacent nodes is set based on the value for the number of consecutive playbacks of instances of content from the viewing behavior;
• generating, by the computing device, a second output using a second prediction network, wherein a second input to the second prediction network is based on a second number of behaviors within a second time threshold that includes a greater amount of time than the first time threshold;
• and
• combining, by the computing device, the first output and the second output to generate a prediction that predicts whether a current behavior is indicative of result in the viewing behavior of the number of consecutive playbacks.