Off-policy evaluation of a new “target” policy is performed using historical data gathered based on a previous “logging” policy to estimate the performance of the target policy. An estimator may be used, wherein either a quality-based estimator or a quality-agnostic estimator is used to weight the difference between an observed reward in the historical data and an estimated reward generated by the target policy. A quality-agnostic estimator may be used to evaluate an importance weight according to a threshold. In such examples, when the importance weight exceeds the threshold, the quality-agnostic estimator clips the importance weight at the threshold, thereby providing an fixed upper bound irrespective of the quality of the reward predictor. In other examples, a quality-based estimator is used, in which an upper bound incorporates the quality of the reward predictor in order to modify an importance weight used by the estimator.

The embodiments set forth techniques for managing advertisement attributions. A first technique can be implemented by an “app store” application, and include the steps of (1) receiving, from a first user application, a request to view a second user application, where the request includes a set of digitally-signed parameters that are specific to an advertising campaign (provided by an advertisement network) for the second user application that is presented by the first user application. In turn, and in response to identifying that the second user application satisfies at least one criterion, the app store application provides the set of digitally-signed parameters to an advertisement metrics manager that: (i) verifies the set of digitally-signed parameters, and (ii) indicates, to the advertisement network, that business logic should be carried out in association with the first user application and the second user application. A second technique for managing advertisement attributions is also disclosed.

A content influencer scoring system may include influencer computers each associated with a respective content influencer having influencer historical performance data and legacy influencer content associated therewith. A remote server may obtain advertisement campaign data associated with an advertisement campaign and parse the advertisement campaign data for advertisement keywords. The remote server may match content influencers to the advertisement campaign data based on the advertisement keywords and, for each content influencer, generate an advertisement campaign score. The score may be generated by determining whether the content influencer is suitable for the advertisement campaign based upon a term frequency of the advertisement keywords for each document from the legacy influencer content, and frequency of the advertisement keywords across the documents, and when suitable, determining whether the advertisement campaign score based upon the historical performance data to generate the advertisement campaign score.

Embodiments of system are disclosed in which selection strategies for a direct marketing campaign that identify consumers from a credit bureau or other consumer database can be planned, tested, and/or refined on a stable subset of the credit database. In some embodiments, once refined, consumer selection criteria may be used to execute the direct marketing campaign on the full consumer/credit database, which is preferably updated approximately twice weekly. In one preferred embodiment, the data for the test database represents a random sampling of approximately 10% of the full database and the sampling is regenerated approximately weekly in order to provide a stable set of data on which campaign developers may test their campaign. For each consumer in the sampling, the environment may allow a client to access and use both attributes calculated by the credit bureau and proprietary attributes and data owned by the client. The system allows for a plurality of clients to use the system substantially simultaneously while protecting the privacy and integrity of the client's proprietary data and results.

Executing a machine learning model in an artificial intelligence infrastructure that includes one or more storage systems and one or more graphical processing unit (‘GPU’) servers, including: receiving, by a graphical processing unit (‘GPU’) server, a dataset transformed by a storage system that is external to the GPU server; and executing, by the GPU server, one or more machine learning algorithms using the transformed dataset as input.

Techniques for determining recommended allocations of resources among different platforms that sell a common type of inventory. Determining the allocations can include obtaining parameters of a campaign from a client. Determining the allocations can include combining current campaign parameters and scoring with historical campaign performance data to create recommendations for dividing resources among different media platforms.

One-to-one digital media modeling systems and methods are disclosed for optimizing digital media reach within digital networks. A data seed is generated that defines a seed audience of users defined by targeting criteria for a digital media asset. The data seed is provided to a lookalike algorithm that applies the data seed to a userbase comprising user data of additional users to generate a lookalike media model comprising a campaign audience dataset defining a plurality of audience datasets having a relevancy score and each having users selected from the seed audience or the additional users. An exposed lookalike audience dataset is created by merging each of the plurality of audience datasets having the relevancy score above a relevancy threshold value, wherein the exposed lookalike audience dataset defines a subset of targeted users. The digital media asset is then transmitted across a digital network for display on a user device.

One or more computing devices, systems, and/or methods for isolated budget utilization are provided. A first budget pacing component is assigned to control bidding by a first content serving component for a set of content items. A second budget pacing component is assigned to control bidding by a second content serving component for the set of content items. The first budget pacing component controls the bidding by the first content serving component according to a first portion of a content item budget based upon a traffic share of the first content serving component. The second budget pacing component controls the bidding by the second content serving component according to a second portion of the content item budget based upon a traffic share of the second content serving component.

Automated methods are provided for identifying when a first advertisement (ad) is a likely match of either a second ad, or one or more variants of the second ad. The first ad, the second ad, and the one or more variants of the second ad each include a plurality of sequential segments of a predefined time length, wherein the second ad and the one or more variants of the second ad are each reference ads, and the first ad is a sample ad. Vectors of segment hits are created for the various ads and are compared to each other to identify matches that represent such variants.

A platform security system and method improve security by binding an identity of a self-contained certificate signing request (SC CSR) requestor to the SC CSR to prevent malicious tampering, such as man-in-the-middle attacks. In at least one embodiment, the requestor, such as a client computer system or other source of a request, requests certificates from a certificate authority (CA). Binding the identity of the SC CSR to the requestor can prevent unauthorized system and/or data access and potentially resultant unauthorized access, malicious tampering, such as man-in-the-middle attacks, and other unauthorized actions or observations. Validation can be performed at the CA on the SC CSR to determine the integrity of the requestor and authorization to receive certificates before the CA sends the certificate to the requestor.