Abstract: Systems and methods are directed to providing multilevel chained testing. A modeling manager receives a request for data associated with an experience having multiple levels of testing, whereby each lower level of testing has a set of one or more variants chained to a variant of a higher level. Based on the request, the model manager determines which variant of the multiple levels of testing to provide to a user. The determining comprises detecting a lowest segment the user is a member of, whereby each segment level corresponds to a level of testing, and selecting a variant from a corresponding set of one or more variants of the lowest sub-segment, a chained variant of a parent segment, or a control value. The modeling manager transmits a response to an experience component that includes the selected variant, and the experience component causes presentation of the experience with the selected variant.

Abstract: Methods, systems, and computer programs are presented for implementing an artificial-intelligence modeling utility system. One method includes receiving, by a modeling manager, a schema from an experience module that implements features of an online service. The modeling manager manages a plurality of machine-learning (ML) models, provides a user interface (UI) based on the schema for entering experiment parameter values, and configures one or more ML models for the experiment. The experiment is initialized, and during the experiment, the modeling manager receives a request from the experience module for data associated with the experiment and selects one of the configured ML models for providing a response to the request. The response is obtained from the selected ML model based on input provided to the ML model based on the request, and the modeling manager sends the response to the experience. Further, results of the experiment are presented.

Abstract: In an example embodiment, a deep machine learning model ranks cohorts of users as well as cohorts of products in a single ranking. When utilized to determine which cohort members to display to a user, the system selects one user cohort and one product cohort as the “best” (e.g., the top ranked user cohort and the top ranked product cohort). This ranking may be based on a number of contextual and non-contextual features, including viewer features (characteristics of the user operating the user interface), viewee features (characteristics of or related to the litem that the user is viewing, such as the characteristics of another user whose profile the user is viewing), and viewer-viewee relationship features (indications about how the viewer and viewee are related, such as common schools, locations, places of employment, etc.).

Abstract: In an example embodiment, machine learning is utilized to make recommendations for next actions by users of an online network. These next actions are called “next best actions.” The machine learning may be performed to train a multitask deep machine learning model to make recommendations based on a series of inputs, including, for example, contextual information that relies upon action sequences of the user and historical users, and user intent. The use of a multitask deep machine learning model allows for the model to generate action recommendations that are personalized, contextual, and coordinate across various different aspects of the online network, rather than being limited to only a single aspect. Likewise, the multi-task deep machine learning model can also be tailored to optimized different use-case specific objectives while at the same time being easy to scale and maintain.

Abstract: Techniques for using machine learning to leverage deep segment embeddings are provided. In one technique, a set of training data is processed using one or more machine learning techniques to train a neural network and learn an embedding for each segment of multiple segments. In response to receiving a request, multiple elements are identified, such as a source entity that is associated with the request, a source embedding for the source entity, a particular segment with which the source entity is associated, a segment embedding for the particular segment, and multiple target entities. For each target entity, a target embedding is identified and the target embedding, the source embedding, and the segment embedding are input into the neural network to generate output that is associated with the target entity. Based on the output, data about a subset of the target entities is presented on a computing device.

Abstract: Techniques for using machine learning to leverage deep segment embeddings are provided. In one technique, a set of training data is processed using one or more machine learning techniques to train a neural network and learn an embedding for each segment of multiple segments. In response to receiving a request, multiple elements are identified, such as a source entity that is associated with the request, a source embedding for the source entity, a particular segment with which the source entity is associated, a segment embedding for the particular segment, and multiple target entities. For each target entity, a target embedding is identified and the target embedding, the source embedding, and the segment embedding are input into the neural network to generate output that is associated with the target entity. Based on the output, data about a subset of the target entities is presented on a computing device.

Abstract: A method, system, and computer program product for of load balancing in a parallel map/reduce paradigm. The method commences by sampling a first set of input records, and forming a prospective load balancing assignment by assigning the first set of input records to the plurality of worker tasks based on a workload estimate for each of the worker tasks. To measure the prospective load balancing assignment, the method compares the workload variance over the plurality of worker tasks to a workload variance target, and also calculates a confidence level based on the sampled first set of input records. If the measured quality of the prospective load balancing assignment is not yet achieved, then the method samples additional input records; for example when the calculated workload variance is greater than the maximum workload variance target or when the calculated confidence level is lower than a confidence level threshold.

Abstract: A method, system, and computer program product for of load balancing in a parallel map/reduce paradigm. The method commences by sampling a first set of input records, and forming a prospective load balancing assignment by assigning the first set of input records to the plurality of worker tasks based on a workload estimate for each of the worker tasks. To measure the prospective load balancing assignment, the method compares the workload variance over the plurality of worker tasks to a workload variance target, and also calculates a confidence level based on the sampled first set of input records. If the measured quality of the prospective load balancing assignment is not yet achieved, then the method samples additional input records; for example when the calculated workload variance is greater than the maximum workload variance target or when the calculated confidence level is lower than a confidence level threshold.