In some embodiments, a method receives data for sessions that is generated based on use of a service. The data for a session includes session characteristic values for a set of session characteristics that is associated with the session. The method classifies the sessions into a plurality of session types to generate feature values for the plurality of session types. The session characteristic values for the session are compared to a set of reference characteristics for respective session types in the plurality of sessions to perform the classifying. The feature values are analyzed for the plurality of session types using a model to generate a prediction for the service and the is outputted prediction.
BACKGROUND
A company may generate predictions from information related to user data. For example, a company may provide a service, such as a video delivery service in which users playback content, and the company may want to generate different predictions, such as whether the user account may become disengaged with the service, upgrade the service, increase or decrease engagement, watch a type of video, etc. In some examples, a company may use a prediction network that can generate the predictions. A typical method may be to determine values for features based on the user account that is using the service. The values are input into a model that is implemented by the prediction network to generate a prediction.
Sometimes, more features are added in an attempt to generate a more accurate prediction. For example, some features may be the result of coarse feature aggregation, such as total hours streamed in the past month, the number of items added to a watchlist for later viewing since signup, etc. These highly aggregated numbers may fail to capture true user account level engagement and interactions with the service. Also, as more features are added, the model may not be able to sensibly analyze all the features to determine a prediction that makes sense, essentially leading to overfitting where the model fits too closely against the training data used for the features, but not for new data in which a prediction is being determined. Also, once the prediction is re
1. A method comprising: receiving, by a computing device, data for sessions that is generated based on use of a service, wherein the data for a session includes session characteristic values for a set of session characteristics that is associated with the session; classifying, by the computing device, the sessions into a plurality of session types, wherein the session characteristic values for the session are compared to a set of reference characteristics for respective session types in the plurality of sessions to perform the classifying; generating, by the computing device feature values based on a number of occurrences for session types in the plurality of session types based on sessions being classified in respective session types; analyzing, by the computing device, the feature values for the plurality of session types using a model to generate a prediction for the service; and outputting, by the computing device, the prediction. ||
17. A non-transitory computer-readable storage medium having stored thereon computer executable instructions, which when executed by a computing device, cause the computing device to be operable for: receiving data for sessions that is generated based on use of a service, wherein the data for a session includes session characteristic values for a set of session characteristics that is associated with the session; classifying the sessions into a plurality of session types, wherein the session characteristic values for the session are compared to a set of reference characteristics for respective session types in the plurality of sessions to perform the classifying; generating feature values based on a number of occurrences for session types in the plurality of session types based on sessions being classified in respective session types; analyzing the feature values for the plurality of session types using a model to generate a prediction for the service; and outputting the prediction. 18.- ||
21. An apparatus comprising: one or more computer processors; and a computer-readable storage medium comprising instructions for controlling the one or more computer processors to be operable for: receiving data for sessions that is generated based on use of a service, wherein the data for a session includes session characteristic values for a set of session characteristics that is associated with the session; classifying the sessions into a plurality of session types, wherein the session characteristic values for the session are compared to a set of reference characteristics for respective session types in the plurality of sessions to perform the classifying; generating feature values based on a number of occurrences for session types in the plurality of session types based on sessions being classified in respective session types; analyzing the feature values for the plurality of session types using a model to generate a prediction for the service; and outputting the prediction.