a method receives a first value for a feature.
In some embodiments, a method receives a first value for a feature. The first value is associated with an entity. A cohort is determined for the entity where a cohort is associated with a dimension, and the entity is associated with a dimension value for the dimension. The method determines a normalization value for the cohort based on the dimension value for the entity. The normalization value is generated based on second values at the dimension value for the feature, and the second values are associated with entities in the cohort. A cohort-agnostic feature value is generated based on the first value and the normalization value. The cohort-agnostic feature value is input into a model to generate a prediction for the entity.
BACKGROUND (1) A service may generate a prediction for a user account of the service, such as a prediction of an action, an outcome, a classification, etc. For example, a content delivery service may want to predict whether a user account will upgrade the service, cancel the service, watch some content, classify the user account in a group (e.g., age groups, gender, parents with kids, single, etc.), etc. Other predictions may be used, such as upgrading a bank account, classification of users that attend parks, etc. Features may be used as input to models to generate the predictions. For example, if a feature of hours streamed is used, a problem with temporality may occur, such as a prediction for a first user account who streamed 10 hours and signed up for the service a week ago may be the same as a second user account who streamed the same number of hours, but signed up a month ago. However, generating the same prediction for the first user account and the second user account may not be an accurate prediction. Indeed, in the previous example, the user account having streamed 10 hours over a week will have a much higher daily engagement than the user account with 10 hours over a month, and the system should therefore expect different predictions for these two user accounts. (2) To address the above problem, multiple models may be used to capture the differences in tenure. For example, one model may be used for user accounts that have a tenure on the service of less than a cer
1. A method comprising: receiving, by a computing device, a first value for a feature, wherein the first value is associated with an entity; determining, by the computing device, a cohort for the entity, wherein a cohort is associated with a dimension, and the entity is associated with a dimension value for the dimension; determining, by the computing device, a normalization value for the cohort based on the dimension value for the entity, wherein the normalization value is generated based on second values at the dimension value for the feature, and wherein the second values are associated with entities in the cohort; generating, by the computing device, a cohort-agnostic feature value based on the first value and the normalization value; and inputting, by the computing device, the cohort-agnostic feature value into a single model to generate a prediction for the entity, wherein the single model is used to process cohort-agnostic feature values from multiple entities that are associated with different dimension values. ||
14. 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 a first value for a feature, wherein the first value is associated with an entity; determining a cohort for the entity, wherein a cohort is associated with a dimension, and the entity is associated with a dimension value for the dimension; determining a normalization value for the cohort based on the dimension value for the entity, wherein the normalization value is generated based on second values at the dimension value for the feature, and wherein the second values are associated with entities in the cohort; generating a cohort-agnostic feature value based on the first value and the normalization value; and inputting the cohort-agnostic feature value into a single model to generate a prediction for the entity, wherein the single model is used to process cohort-agnostic feature values from multiple entities that are associated with different dimension values. ||
20. 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 a first value for a feature, wherein the first value is associated with an entity; determining a cohort for the entity, wherein a cohort is associated with a dimension, and the entity is associated with a dimension value for the dimension; determining a normalization value for the cohort based on the dimension value for the entity, wherein the normalization value is generated based on second values at the dimension value for the feature, and wherein the second values are associated with entities in the cohort; generating a cohort-agnostic feature value based on the first value and the normalization value; and inputting the cohort-agnostic feature value into a single model to generate a prediction for the entity, wherein the single model is used to process cohort-agnostic feature values from multiple entities that are associated with different dimension values.