Google Inc.

Prachi Gupta

Eyal Manor

Thomas Price

Sarvjeet Singh

Deepak Ramamurthi Sivaramapuram Chandrasekaran

Jierui Xie

Bo Fu

Sue Yi Chew

Kunal Jain

A request from a user to view a video content item may be received, the requesting user being associated with at least one of a set of preferences or a context. A group of similar users may be identified based the set of preferences or the context. A number of candidate video items corresponding to the video content item may be identified. A watch time difference may be determined for each candidate video item of the number of candidate video items. A subset may be determined based on the watch time difference associated with each candidate video items. The candidate video items in the subset may be ranked based on an activity rate associated with the corresponding candidate video item. A candidate video item with the highest ranking may be provided to the user.

Providing a message based on a change in watch time

Balancing content distribution between a machine learning model and a statistical model provides a baseline assurance in combination with the benefits of a well-trained machine learning model for content selection. In some implementations, a server receiving requests for a content item assigns a first proportion of the received requests to a first group and assigns remaining requests to a second group. The server uses a machine learning model to select variations of the requested content item for responding to requests assigned to the first group and uses a statistical model to select content variations for requests assigned to the second group. The server obtains performance information, e.g., acceptance rates for the different variations, and compares performance of the different models used for content selection. Audience share assigned to the machine learning model is increased when it outperforms the statistical model and decreased when it underperforms the statistical model.

Assessing accuracy of a machine learning model

A request from a user to view a video content item may be received, the requesting user being associated with a set of preferences and a context. A group of similar users may be identified based the set of preferences or the context. A number of promotional video items corresponding to the video content item may be identified. A first subset of the number of promotional video items may be determined based on the set of preferences or the context of the user. A watch time difference may be determined for each promotional video item in the first subset. A second subset may be determined based on the watch time difference associated with each promotional video items. An activity rate associated with the promotional video items in the second subset is determined. A promotional video item of the second subset that satisfies a criterion is provided to the user.

A request from a user to view a video content item may be received, the requesting user being associated with a set of preferences and a context. A group of similar users may be identified based at least one of: one or more preferences in the set of preferences or the context. A first promotional video item corresponding to the video content item may be identified. A first watch time value for a first portion of the group of similar users may be determined based on an amount of time during a period of time that users of the first portion of the group have viewed the video content item after being provided the first promotional video item. A second watch time value for a second portion of the group of similar users may be determined based on an amount of time during the period of time that the users of the second portion of the group have viewed the video content item, the first promotional video item not provided to the users of the second portion of the group. A watch time difference between the first watch time value and the second watch time value may be determined. An activity rate associated with the first promotional video item that was provided to the group of similar users may be determined. A determination of whether to provide the first promotional video item to the user with the video content item may be based on the watch time difference, the activity rate, or the context of the user.

A promotional message may be provided to a first portion of a group of users. A first watch time may be identified based on an amount of time during a period of time that users of the first portion view the video content items after being provided the promotional message. A second watch time may be identified based on an amount of time during the period of time that users of a second portion of the group view video content items where the promotional message is not provided to the users of the second portion of the group. A difference between the first watch time and the second watch time may be calculated. A determination of whether to provide the promotional message to a subsequent user sharing one or more characteristics with the group of users may be based on the difference between the first watch time and the second watch time may be made.

Growing community of inventors

Los Altos, CA, United States of America

Prachi Gupta