Abstract: Techniques for varying content item density are provided. A first minimum gap value is stored that dictates how close two content items of a first type may appear in a content item feed that contains content items of multiple types that includes the first type and a second type. The first minimum gap value is used to place content items in a first set of content item feeds. For each content item feed of the first set of content item feeds, performance data that indicates how well content items of the first type perform in the content item feed is generated. Based on the performance data and the first minimum gap value, a second minimum gap value that is different than the first minimum gap value is generated. The second minimum gap value is used to place content items in a second plurality of content item feeds.

Abstract: Incremental model training for advertisement targeting is performed using streaming data. A model for targeting advertisements of an advertising campaign is initialized. A data stream including data corresponding to converters and data corresponding to non-converters is received. The model is then applied to the data corresponding to the converter and data corresponding to the non-converter (or other ratio of converter to non-converters) to obtain a predicted score for each. The predicted score is compared to the observed score (e.g., an observed score of 1 for a converter, and 0 for a non-converter). The difference between the predicted and observed scores is computed, and the model is incrementally updated based on this difference. Models can optionally be built separately on multiple modeling servers that are geographically dispersed in order to support bidding on advertising opportunities in a real-time bidding environment.

Abstract: Technologies for associating an entity with a content delivery campaign are provided. Disclosed techniques include determining a first value of a profile attribute of the entity. A particular node that matches the first value is identified from a value tree of nodes. A parent node of the particular node is identified from the value tree. Child nodes of the parent node are identified, where the child nodes do not include the particular node. Values from the child nodes are then associated with the profile attribute of the entity. A particular value is received for a particular targeting criterion of the content delivery campaign. It is determined whether the particular value matches a value of the child nodes, where the particular value does not match the first value. In response to determining that the particular value matches a value of the child nodes, associating the entity with the content delivery campaign.

Abstract: Techniques for automatically merging multiple content item queues are provided. In one technique, a first set of content items of a first type is identified. A second set of content items of a second type that is different than first type is identified. The first set of content items and the second set of content items are merged in a content item feed. Such merging involves, for a particular slot in the content item feed: determining a previous slot that contains a first content item from the first set; determining a number of slots between the previous slot and the particular slot; based on the number of slots, generating a score for a second content item from the second set; and based on the score, determining whether to insert, into the particular slot, the second content item or a third content item from the first set of content items.

Abstract: Techniques for varying content item density are provided. A first minimum gap value is stored that dictates how close two content items of a first type may appear in a content item feed that contains content items of multiple types that includes the first type and a second type. The first minimum gap value is used to place content items in a first set of content item feeds. For each content item feed of the first set of content item feeds, performance data that indicates how well content items of the first type perform in the content item feed is generated. Based on the performance data and the first minimum gap value, a second minimum gap value that is different than the first minimum gap value is generated. The second minimum gap value is used to place content items in a second plurality of content item feeds.

Abstract: Techniques for automatically merging multiple content item queues are provided. In one technique, a first set of content items of a first type is identified. A second set of content items of a second type that is different than first type is identified. The the first set of content items and the second set of content items are merged in a content item feed. Such merging involves, for a particular slot in the content item feed: determining a previous slot that contains a first content item from the first set; determining a number of slots between the previous slot and the particular slot; based on the number of slots, generating a score for a second content item from the second set; and based on the score, determining whether to insert, into the particular slot, the second content item or a third content item from the first set of content items.

Abstract: Techniques for reducing delay in broadcasting content over a network using an inverted fan-out process are disclosed herein. In some embodiments, a computer-implemented method comprises: in response to an activity associated with content being performed by a user on an online service, detecting that the activity has been performed: identifying a plurality of recipient users in response to the detecting; and for each one of the plurality of recipient users, transmitting a notification of the activity to a destination associated with the recipient user in response to the identifying of the recipient users, the notification comprising an indication of the content, and the transmitting of the notification of the activity being performed without waiting for the recipient user to navigate to a web page of the online service on a computing device or for the recipient to open a mobile application of the online service on a mobile device.

Abstract: A system comprises servers comprising processors and memory storing instructions which, when executed by the processors, cause the processors to perform: receiving, from a content provider, a request to begin a content delivery campaign, the request comprising targeting criteria and a resource amount, prior to beginning the content delivery campaign, determining, based on the targeting criteria, historical results of content item selection events, prior to beginning the content delivery campaign, determining, based on the targeting criteria, a suggested resource reduction request amount using the historical results and the resource amount, causing the suggested resource reduction request amount to be displayed to the content provider, receiving, from the content provider, a particular resource reduction request amount for the content delivery campaign, initiating the content delivery campaign using the particular resource reduction request amount.

Abstract: Incremental model training for advertisement targeting is performed using streaming data. A model for targeting advertisements of an advertising campaign is initialized. A data stream including data corresponding to converters and data corresponding to non-converters is received. The model is then applied to the data corresponding to the converter and data corresponding to the non-converter (or other ratio of converter to non-converters) to obtain a predicted score for each. The predicted score is compared to the observed score (e.g., an observed score of 1 for a converter, and 0 for a non-converter). The difference between the predicted and observed scores is computed, and the model is incrementally updated based on this difference. Models can optionally be built separately on multiple modeling servers that are geographically dispersed in order to support bidding on advertising opportunities in a real-time bidding environment.

Abstract: Incremental model training for advertisement targeting is performed using streaming data. A model for targeting advertisements of an advertising campaign is initialized. A data stream including data corresponding to converters and data corresponding to non-converters is received. The model is then applied to the data corresponding to the converter and data corresponding to the non-converter (or other ratio of converter to non-converters) to obtain a predicted score for each. The predicted score is compared to the observed score (e.g., an observed score of 1 for a converter, and 0 for a non-converter). The difference between the predicted and observed scores is computed, and the model is incrementally updated based on this difference. Models can optionally be built separately on multiple modeling servers that are geographically dispersed in order to support bidding on advertising opportunities in a real-time bidding environment.

Abstract: A method for performance bottleneck diagnosis and dependency discovery in distributed systems and computer networks includes receiving a real-valued end-to-end measurement matrix, a number of end-to-end measurements, a number of time points, a number of network components, a loss function, and a plurality of constraints on output matrices. The method further includes learning basic output matrices by best approximation of a transpose of the real-valued end-to-end measurement matrix, selecting a first threshold based on a real-valued mixing-weights matrix and a second threshold based on a real-valued delay matrix, converting the real-valued mixing-weights matrix and the real-valued delay matrix into respective binary matrices using the first threshold and the second threshold, and returning the real-valued mixing-weights matrix, the real-valued delay matrix, and the respective binary matrices to represent bottlenecks dependencies of the distributed system or computer network.

Abstract: A method for performance bottleneck diagnosis and dependency discovery in distributed systems and computer networks includes receiving a real-valued end-to-end measurement matrix, a number of end-to-end measurements, a number of time points, a number of network components, a loss function, and a plurality of constraints on output matrices. The method further includes learning basic output matrices by best approximation of a transpose of the real-valued end-to-end measurement matrix, selecting a first threshold based on a real-valued mixing-weights matrix and a second threshold based on a real-valued delay matrix, converting the real-valued mixing-weights matrix and the real-valued delay matrix into respective binary matrices using the first threshold and the second threshold, and returning the real-valued mixing-weights matrix, the real-valued delay matrix, and the respective binary matrices to represent bottlenecks dependencies of the distributed system or computer network.