Abstract: A computer system constructs a decision-predictive recipient profile using sensatory data tied to an online profile of a recipient. After obtaining base sensatory data tied to the online profile of the recipient, the system may filter the base sensatory data by searching the base sensatory data for one or more machine-cognizable characteristics. The filtered sensatory data may be provided to an execution group, which may review displays of the sensatory data. Responsive to the displays of the sensatory data, the execution group may generate descriptors of the content of the filtered sensatory data and send the descriptors to the system. The system may process the descriptors to generate or augment the decision-predictive recipient profile.

Abstract: According to examples, intelligent visualization munging may include transforming and enriching data that is to be visualized, determining features of the transformed and enriched data, determining a user role of a user associated with the transformed and enriched data, and a user interaction of the user. Intelligent visualization munging may further include learning a behavior of the user, and analyzing the features, the user role, the user interaction, and a learned behavior model to generate a recommendation that includes a predetermined number of visualizations from a plurality of available visualizations to display the transformed and enriched data. The predetermined number of visualizations is less than the plurality of available visualizations.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for centralized management of predictive models. One of the methods includes receiving a plurality of model versions from respective teams within the organization. Each received model version is processed to generate a respective predictive model corresponding to the model version, including identifying one or more collections of training data specified by the model version, identifying one or more training engines specified by the model version, training, by the identified one or more training engines, a predictive model using the identified one or more collections of training data, computing one or more performance metrics for the predictive model, and associating the one or more performance metrics for the predictive model with the corresponding model version in the version repository.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for forecasting event volumes. One of the methods includes receiving activity data associated with events that occur in a computer network, transforming the received activity data into processed activity data, determining, for each of a series of distinct time periods, whether the distinct time period is associated with an anomalous level of event activity, and flagging the processed activity data with a designation of whether each of the series of distinct time periods is associated with the anomalous level of event activity. Based on the flagged activity data, prediction data is generated and provided that predicts a volume of events for a given future time period and a likelihood that the volume of events for the given future time period will be anomalous.

Abstract: A computer system constructs a decision-predictive recipient profile using sensatory data tied to an online profile of a recipient. After obtaining base sensatory data tied to the online profile of the recipient, the system may filter the base sensatory data by searching the base sensatory data for one or more machine-cognizable characteristics. The filtered sensatory data may be provided to an execution group, which may review displays of the sensatory data. Responsive to the displays of the sensatory data, the execution group may generate descriptors of the content of the filtered sensatory data and send the descriptors to the system. The system may process the descriptors to generate or augment the decision-predictive recipient profile.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for centralized management of predictive models. One of the methods includes receiving a plurality of model versions from respective teams within the organization. Each received model version is processed to generate a respective predictive model corresponding to the model version, including identifying one or more collections of training data specified by the model version, identifying one or more training engines specified by the model version, training, by the identified one or more training engines, a predictive model using the identified one or more collections of training data, computing one or more performance metrics for the predictive model, and associating the one or more performance metrics for the predictive model with the corresponding model version in the version repository.

Abstract: According to examples, intelligent visualization munging may include transforming and enriching data that is to be visualized, determining features of the transformed and enriched data, determining a user role of a user associated with the transformed and enriched data, and a user interaction of the user. Intelligent visualization munging may further include learning a behavior of the user, and analyzing the features, the user role, the user interaction, and a learned behavior model to generate a recommendation that includes a predetermined number of visualizations from a plurality of available visualizations to display the transformed and enriched data. The predetermined number of visualizations is less than the plurality of available visualizations.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for forecasting event volumes. One of the methods includes receiving activity data associated with events that occur in a computer network, transforming the received activity data into processed activity data, determining, for each of a series of distinct time periods, whether the distinct time period is associated with an anomalous level of event activity, and flagging the processed activity data with a designation of whether each of the series of distinct time periods is associated with the anomalous level of event activity. Based on the flagged activity data, prediction data is generated and provided that predicts a volume of events for a given future time period and a likelihood that the volume of events for the given future time period will be anomalous.