Hulu, LLC

Chunyang Wei

Dingming Wu

Shulei Ma

Xiaohui Xie

Changsheng Xiang

Pengfei Gao

Qingying Chen

Wu Zhang

Hanqi Wang

In some embodiments, a method receives a line item factor for a line item at a first model. The method receives a competition factor at a second model. The first model generates a first value for a representation parameter based on the line item factor. The representation parameter is used to generate a representation for estimating selection of the instance of supplemental content for the line item for delivery. The second model generates a modification factor based on the competition factor. The modification factor is configured to modify the first value for the representation parameter based on competition from other line items to select the instance of supplemental content for the line item for delivery. The method modifies the first value for the representation parameter using the modification factor to generate a second value for the representation parameter where the second value is used to generate the representation.

Representation parameter adjustment using separate models

In some embodiments, a method determines a reward metric based on feedback for an instance of content. A delivery status for a delivery constraint of the instance of content is applied to the reward metric to generate a constrained reward metric. The method uses the constrained reward metric to train a model. The model is used to select from a plurality of instances of content. One of the plurality of instances of content is selected for delivery using the model.

Reinforcement learning for delivery of content

In some embodiments, a method receives information for a delivery of instances of supplemental content for a plurality of line items. A line item is associated with an instance of supplemental content that can be delivered and a pacing curve that describes a pace of delivery over time. The method updates a parameter for the line item to generate an updated parameter based on the delivery of the instances of supplemental content and a desired pacing behavior. The updated parameter is provided to a selection system that uses the updated parameter to select an instance for delivery. The delivery of instances of supplemental content for the line item is adjusted to meet the pacing curve based on a characteristic of the pacing behavior.

Optimal supplemental content selection in content delivery

In some embodiments, a method receives a first set of probabilities mapping a behavior to a user for a set of behaviors in a video delivery system. A first part of a model is used to predict the first set of probabilities. The method receives a second set of probabilities mapping a user to a behavior for a set of users that are using an account on the video delivery system. A second part of the model is used to predict the second set of probabilities. A combined probability is calculated for the set of behaviors and the set of users based on the first set of probabilities and the second set of probabilities. The method updates a second parameter of the second part of the model while keeping a first parameter of the first part of the model fixed and outputs the second parameter of the second part of the model upon determining the training of the model is finished.

Training of multiple parts of a model to identify behavior to person prediction

In some embodiments, a method inputs a sequence of historical behaviors for a plurality of instances of content into a prediction network to generate a sequence of values that model the sequence of historical behaviors. A restriction on an operation performed by the prediction network is based on a characteristic of an viewing behavior. A sequence of attention scores is generated 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. The method adjusts respective values based on corresponding attention scores to generate an adjusted sequence of values. The adjusted sequence of features are sampled 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.

Account behavior prediction using prediction network

In some embodiments, a method receives an supplemental content placement and a context associated with a request for supplemental content to be displayed for the supplemental content placement. A first value is generated based on the context using a prediction network for a platform. The method determines probabilities for a plurality of types of request actions based on the context. Then, a threshold for the supplemental content placement is calculated based on the first value and the probabilities for the plurality of types of request actions. The method submits the threshold to a platform in a request for the platform to submit a second value for the supplemental content placement.

Dynamic threshold calculation for video streaming

In some embodiments, a method receives a request for supplemental content to be provided in association with main content. The method selects an instance of supplemental content based on a long-term reward metric and a short-term reward metric. The long-term reward metric is based on feedback from delivery of a plurality of instances of supplemental content and a delivery status for a delivery constraint of one instance of supplemental content. The short-term reward metric is based on feedback from delivery of the one instance of supplemental content. The long-term reward metric is based on feedback from delivery of a plurality of instances of supplemental content and the short-term reward metric is based on feedback from delivery of one instance of supplemental content. The instance of supplemental content is provided to a client device.

Growing community of inventors

Beijing, China

Chunyang Wei