Yahoo Ad Tech LLC

Yahoo Assets LLC

Verizon Media Inc.

Yohay Kaplan

Yair Koren

Oren Shlomo Somekh

Tal Cohen

Alexander Shtof

Roie Melamed

Abraham Shahar

Alexander Zlotnik

Naama Haramaty-Krasne

Alexander Shtoff

Michal Aharon

Lakshmi Narayan Bhamidipati

Rotem Stram

Tomer Shadi

Anna Itzhaki

Boris Trayvas

Ravi Kant

Avi Shahar

Alex Shtoff

The present teaching relates to displaying ads. An explore/exploit layer (EEL) is provided at frontend ad serving engine for storing combination distributions with respect to multiple ads. Each ad has multiple attributes. Each attribute can be instantiated using one of multiple assets. The frontend ad serving engine requests a recommended ad for bidding an ad display opportunity in a slot of a webpage viewed by a user on a user device. The recommended ad is one of the multiple ads. When the auction is successful, a combination of assets for the ad is drawn from the combination distributions in EEL and each of the assets instantiates a corresponding attribute of the ad. The combination is transmitted to the user device to render the ad.

System and method for thin explore/exploit layer for providing additional degree of freedom in recommendations

One or more computing devices, systems, and/or methods for content recommendations using historical future data are provided. A model serving delay time is computed as an average of training delays of events. A historical data time interval is determined based upon the model serving delay time. A model is trained for predicting user content preferences using historic user distribution data and historic content distribution data associated with the historic data time interval. The model is utilized to generate and provide content recommendations to users.

Content recommendations based upon historical future data

One or more computing devices, systems, and/or methods for content recommendation based upon continuity and grouping information of attributes are provided herein. User interaction data specifying whether users interacted with content items, user attributes of the users, and content attributes of the content items is obtained. A data structure is populated with the user interaction data. The data structure is modified by inserting a set of sub-fields into the data structure for a user attribute. A sub-field is populated with a value representing an option of the user attribute. The set of sub-fields are an encoding of continuity information and grouping information representing options for the user attribute. The data structure is processed using machine learning functionality to generate a model. The model is utilized to generate a prediction as to whether a user will interact with a content item.

Content recommendation based upon continuity and grouping information of attributes

One or more computing devices, systems, and/or methods for implementing a model for serving exploration traffic are provided. An amount of spend by a content provider to provide content items of the content provider through a content serving platform to client devices of users is determined. A number of exploration impressions of users viewing exploration content items of the content provider over a timespan is determined. A return on exploration impression metric is determined for the content provider based upon a ratio of the amount of spend to the number of exploration impressions. The return on exploration metric is used to rank available exploration content items of content providers for serving exploration traffic.

Model for serving exploration traffic

In an example, sets of event information associated with events may be identified. The events may include intentional click events, accidental click events and/or skip events. Accidental click probabilities associated with the accidental click events and/or the skip events may be determined. Machine learning model training may be performed, using the sets of event information associated with the events and labels associated with the events, to generate a first machine learning model. The labels may include second labels associated with the intentional click events and/or third labels associated with the accidental click events and/or the skip events. The second labels may correspond to an intentional click classification. The third labels may be based upon the accidental click probabilities. Click probabilities associated with content items may be determined using the first machine learning model. A content item may be selected for presentation via a client device based upon the click probabilities.

Content item selection and click probability determination based upon accidental click events

One or more computing devices, systems, and/or methods for selecting content items for presentation via client devices are provided. A content event associated with a content item performed by a client device may be detected. The content item may be associated with an entity. A conversion event, associated with the entity, performed by the client device may be detected. A duration of time between the content event and the conversion event may be determined. An attribution score may be determined based upon the duration of time. A plurality of attribution scores, comprising the attribution score, may be stored in an attribution data structure associated with the content item. Responsive to receiving a request for content associated with a second client device, the content item may be selected from a plurality of content items for presentation via the second client device based upon the attribution data structure.

Content item selection

One or more computing devices, systems, and/or methods for aggregated cost per action prediction are provided. Cost and conversion count data, comprising costs for a set of content items to be displayed to users and a count of conversions corresponding to actions performed by users in response to being provided with the set of content items, is tracked. The cost and conversion count data is inputted into a set of decay calculators that utilize different decay strategies. Cost per action predictions by the set of decay calculators for a content item are weighted to create an aggregated cost per action prediction, wherein the weights are based upon whether decay calculators correctly or incorrectly predicted cost per actions for the set of content items.

Aggregated cost per action prediction

Growing community of inventors

Tel Aviv, Israel

Yohay Kaplan