Abstract: A method including receiving first and second natural language texts. A distance metric is generated from the first and second natural language texts. A first machine learning system is executed, the first machine learning system taking, as a first input, the distance metric and generating, as a first output, a first probability that the first natural language text matches the second natural language text. A second machine learning system is executed, the second machine learning system taking as a second input the first natural language text and as a third input the second natural language text, and generating, as a second output, a second probability that the first natural language text matches the second natural language text. A third probability that the first natural language text matches the second natural language text is generated. Generating includes combining the first probability and the second probability.

Abstract: A method of improving a contextual bandits machine learning model. A training controller is applied to an untrained model, which includes a contextual bandits machine learning model, and training data to generate a trained model. Applying the training controller includes an iterative process that is repeated until convergence. The iterative process includes executing the untrained model on the training data, which includes both privileged information and non-privileged information, to generate a prediction output. A reward is determined, and an updated model is generated based on the reward. The training data, which includes the non-privileged information and excludes the privileged information, is applied to the updated model to generate a test output. A comparison of the test output and the prediction output is used to determine whether convergence has been achieved. The updated model includes the trained model when convergence is achieved.

Abstract: The present disclosure provides techniques for recommending targeted information. One example method includes receiving user data indicative of one or more attributes of one or more users and activity data indicating past actions by the one or more users in association with particular content items, identifying, using a first machine learning model, a topic based on the activity data, identifying, using a second machine learning model, a subset of attributes of the one or more attributes of the one or more users that are associated with the topic, generating a prompt based on the topic and the subset of attributes associated with the topic, and generating, based on the prompt using a large language model (LLM), content to provide to a user having the subset of attributes associated with the topic.

Abstract: Aspects of the present disclosure provide techniques for automated categorization of electronic information. Embodiments include providing inputs to a machine learning model based on attributes of an electronic data item. Embodiments include receiving one or more first outputs from the machine learning model based on the inputs. Embodiments include selecting, based on the one or more first outputs, a question from a plurality of questions. Embodiments include providing the question for display via a user interface. Embodiments include receiving an answer to the question via the user interface. Embodiments include providing updated inputs to the machine learning model based on the answer. Embodiments include receiving one or more second outputs from the machine learning model based on the updated inputs. Embodiments include determining a category for the electronic data item based on the one or more second outputs.

Abstract: Aspects of the present disclosure provide techniques for multi-head machine learning model training. Embodiments include receiving training data comprising training inputs associated with ground truth labels corresponding to a plurality of variables, wherein the ground truth labels include a null value for a given variable of the plurality of variables. Embodiments include providing the training inputs to a machine learning model that is configured to generate predictions corresponding to the plurality of variables. Embodiments include receiving the predictions from the machine learning model in response to the training inputs. Embodiments include evaluating a loss function that compares the ground truth labels to the predictions and uses a masking value to disregard loss that corresponds to the given variable. Embodiments include updating one or more parameters of the machine learning model based on the evaluating of the loss function.

Abstract: A system and method for generating and optimizing marketing campaigns. More specifically, a campaign management system leverages a Large Language Model (LLM) to create multiple variations of an existing campaign tailored to specific target groups. The system employs a cluster-based approach and a click-through rate (CTR) prediction model to generate revised campaigns for targeted readers, thereby creating a feedback loop for further fine-tuning of the LLM for future campaigns.

Abstract: A system for providing content-based (e.g., textual-based) recommendations. The system constructs a knowledge graph from campaign content data including nodes representing individual campaigns and edge weights representing text similarity and collaborative consumption between campaigns. The system adjusts the edge weights within the knowledge graph based on the collaborative consumption of customer groups to emphasize common keywords belonging to common customer groups and deemphasize keywords belonging to different customer groups. The system processes a new campaign content to align with interests of a closest customer group of the customer groups identified in the knowledge graph, thereby enabling targeted delivery of the new campaign to customers associated with that group.

Abstract: Certain aspects of the present disclosure provide techniques for delivering singularly adaptive digital content that includes content components which are adapted in real time based on user interest metrics and using a generative model. Multi-layered content is generated, such that the content may be divided into content components which may each be adapted to be of a selected content type for a content class. The content components include one or more classes having one or more types. To adapt the content, subsequent content components may be adapted by changing which layer of the content component is selected or presented. The content component layers may have been previously generated using a generative model by using a base content component and a selection of content types for content classes. Changing layers for content components may occur when an attention score for the content falls below a threshold to improve interest in the content.

Abstract: Systems and methods for a tunable synchronization network configured to facilitate communication between recommendation systems and generative artificial intelligence systems to generate custom content that is unique to each user. The systems and methods leverage a novel framework including a natural language processing transformer that includes a synchronization layer that can trained to generate profile vectors and a vector translator configured to translate intensity values from profile vectors cells to generate prompts for generative artificial intelligence systems. Notably, this novel framework is agnostic to both the recommendation systems and generative artificial intelligence systems that are used.

Abstract: The present disclosure provides techniques for friendship-based automated recommender system. One example method includes receiving electronic record data indicating interactions between a plurality of users and a plurality of providers, constructing a bipartite graph based on the interactions, identifying, for each user of the plurality of users, a set of other users in the plurality of users, adding to the bipartite graph, for each user of the plurality of users, intra-user edges between the user and the set of other users, computing, for each respective intra-user edge of the intra-user edges, a weight of the respective intra-user edge, computing, for each respective user of the plurality of users, a popularity score, computing, for each respective provider of the plurality of providers, a reputation score, and training a recommender system using the reputation scores of the plurality of providers, wherein the recommender system is used to automatically determine a provider recommendation.

Abstract: A method implements influencer segmentation detection. The method includes selecting transaction data for a time window and processing the transaction data for the time window to generate a graph for the time window. The method further includes extracting, from the graph, a feature set for a node of the graph for the time window and processing the feature set to generate a predicted rank for the node for a subsequent time window using a machine learning model. The method further includes selecting, using the predicted rank, an entity identifier corresponding to the node and presenting the entity identifier.

Abstract: Certain aspects of the present disclosure provide techniques for generating a metric, include receiving a rule defining one or more text strings; determining a set of transactions based on a user attribute; determining a first subset of transactions; determining a second subset of transactions; generating a first categorical distribution based on each transaction of the first subset of transactions being associated with a transaction description containing at least one text string of the one or more text strings; calculating a first unity metric based on the first categorical distribution; generating a second categorical distribution based on each transaction of the second subset of transactions being associated with a transaction description that does not contain a text string of the one or more text strings; calculating a second unity metric based on the second categorical distribution; determining a reliability metric for the rule; and providing the reliability metric.

Abstract: The present disclosure provides techniques for detecting and correcting outliers in categories of transactions. One example method includes receiving electronic transaction data indicative of one or more current transactions, wherein the one or more current transactions are associated with a user of a software application, identifying, for each transaction of the one or more transactions, a category using a first machine learning model, computing a distribution for each category of a plurality of categories of the user, identifying, a particular category of the user as an anomalous category, based on the distribution for the particular category of the user and corresponding distributions for the particular category of other users, and updating a category assigned to one or more transactions such that a delta between a value relating to the anomalous category of the user and corresponding values relating to the particular category of the other users is reduced.

Abstract: The present disclosure provides techniques for recommending vendors using machine learning models. One example method includes receiving electronic transaction data indicative of one or more transactions, identifying, from the one or more transactions, a subset of transactions that are associated with for known attribute values with respect to one or more unique recipients, computing, for each unique provider of the one or more unique providers, a provider feature based on the known attribute values with respect to a subset of the one or more associated unique recipients, computing, for a given recipient indicated in one or more given transactions that are not included in the subset of transactions, a recipient feature based on the provider feature of each unique provider of the one or more associated unique providers, and predicting, based on the recipient feature, a value for the attribute with respect to the given recipient.

Abstract: A method implements influencer segmentation detection. The method includes selecting transaction data for a time window and processing the transaction data for the time window to generate a graph for the time window. The method further includes extracting, from the graph, a feature set for a node of the graph for the time window and processing the feature set to generate a predicted rank for the node for a subsequent time window using a machine learning model. The method further includes selecting, using the predicted rank, an entity identifier corresponding to the node and presenting the entity identifier.

Abstract: Matching validation includes obtaining a candidate match between a target entity and a candidate application user and filtering multiple transaction records of multiple application users to obtain a subset of the transaction records each involving a transaction with the target entity. The application users exclude the candidate application user. Matching validation further includes determining, for each transaction record in the subset, whether a matching transaction record exists in multiple candidate users transaction records of the candidate application user, and validating the candidate match when at least a threshold amount of transaction records in the subset has the matching transaction record in the candidate users transaction records.

Abstract: A method including receiving first and second natural language texts. A distance metric is generated from the first and second natural language texts. A first machine learning system is executed, the first machine learning system taking, as a first input, the distance metric and generating, as a first output, a first probability that the first natural language text matches the second natural language text. A second machine learning system is executed, the second machine learning system taking as a second input the first natural language text and as a third input the second natural language text, and generating, as a second output, a second probability that the first natural language text matches the second natural language text. A third probability that the first natural language text matches the second natural language text is generated. Generating includes combining the first probability and the second probability.

Abstract: Aspects of the present disclosure provide techniques for automated categorization of electronic information. Embodiments include providing inputs to a machine learning model based on attributes of an electronic data item. Embodiments include receiving one or more first outputs from the machine learning model based on the inputs. Embodiments include selecting, based on the one or more first outputs, a question from a plurality of questions. Embodiments include providing the question for display via a user interface. Embodiments include receiving an answer to the question via the user interface. Embodiments include providing updated inputs to the machine learning model based on the answer. Embodiments include receiving one or more second outputs from the machine learning model based on the updated inputs. Embodiments include determining a category for the electronic data item based on the one or more second outputs.

Abstract: Techniques for adaptive validation and remediation are described. In some embodiments, the method includes determining, for a plurality of media service accounts, corresponding fraud suspicion values based on a model. The method also includes identifying a plurality of suspected accounts based on the corresponding fraud suspicion values. The method additionally includes identifying one or more suspected devices and predicting a likelihood of account takeover from each of the one or more suspected devices. The method further includes detecting a triggering event from a device of the one or more suspected devices associated with an account. The method additionally includes executing a validation and/or remediation procedure based on a trigger sensitivity value associated with the triggering event, a respective likelihood of account takeover from the device associated with the account, a respective device risk value associated with the device, and a respective fraud suspicion value associated with the account.

Abstract: Certain aspects of the present disclosure provide techniques for generating a metric, include receiving a rule defining one or more text strings; determining a set of transactions based on a user attribute; determining a first subset of transactions; determining a second subset of transactions; generating a first categorical distribution based on each transaction of the first subset of transactions being associated with a transaction description containing at least one text string of the one or more text strings; calculating a first unity metric based on the first categorical distribution; generating a second categorical distribution based on each transaction of the second subset of transactions being associated with a transaction description that does not contain a text string of the one or more text strings; calculating a second unity metric based on the second categorical distribution; determining a reliability metric for the rule; and providing the reliability metric.