Abstract: Certain aspects of the present disclosure provide techniques for training and using predictive models to predict the occurrence of an event within a software application. An example method generally generating a spatially sampled data set for a set of users of a software application. The spatially sampled data set includes, for each respective user of the set of users, an amount of time the user has spent, a number of discrete portions of the software application the user has visited, and an indication of whether the user has completed a defined task. A spatio-temporally sampled data set for users in the spatially sampled data set is generated, including, for each respective user in the spatially sampled data set, a plurality of candidate timestamps. A predictive model is trained based on the spatio-temporally sampled data set.

Abstract: Certain aspects of the present disclosure provide techniques for operation of a feature management platform. A feature management platform is an end-to-end platform developed to manage the full lifecycle of data features. For example, to create a stateful feature, the feature management platform can receive a processing artifact from a computing device. The processing artifact defines the stateful feature, including the data source to retrieve event data from, when to retrieve the event data, the type of transform to apply, etc. Based on the processing artifact, the feature management system generates a processing job (e.g., the API defines a pipeline), which when initiated generates a vector that encapsulates the stateful feature. The vector is transmitted to the computing device that locally hosts a model, which generates a prediction that is transmitted to the feature management platform. Subsequently, the predication and stateful feature can be transmitted to other computing devices.

Abstract: Certain aspects of the present disclosure provide techniques for training and using predictive models to predict the occurrence of an event within a software application. An example method generally generating a spatially sampled data set for a set of users of a software application. The spatially sampled data set includes, for each respective user of the set of users, an amount of time the user has spent, a number of discrete portions of the software application the user has visited, and an indication of whether the user has completed a defined task. A spatio-temporally sampled data set for users in the spatially sampled data set is generated, including, for each respective user in the spatially sampled data set, a plurality of candidate timestamps. A predictive model is trained based on the spatio-temporally sampled data set.

Abstract: A method including extracting, from an input, supported data. The input includes outputs from machine learning models in different formats. The supported data includes a subset of the input after data normalization. The method also includes inferring, from the supported data, data types to be used with respect to generating metrics for the machine learning models. The method also includes generating, from the supported data and using the data types, a relational event including the supported data. The relational event further includes a first data structure object including the types and having a first data structure different than the different formats. The method also includes calculating, using the supported data in the first data structure, the metrics for the machine learning models. The method also includes generating, from the relational event, a monitoring event. The monitoring event includes a second data structure object segmented into data buckets storing the metrics.

Abstract: Certain aspects of the present disclosure provide techniques for operation of a feature management platform. A feature management platform is an end-to-end platform developed to manage the full lifecycle of data features. For example, to create a stateful feature, the feature management platform can receive a processing artifact from a computing device. The processing artifact defines the stateful feature, including the data source to retrieve event data from, when to retrieve the event data, the type of transform to apply, etc. Based on the processing artifact, the feature management system generates a processing job (e.g., the API defines a pipeline), which when initiated generates a vector that encapsulates the stateful feature. The vector is transmitted to the computing device that locally hosts a model, which generates a prediction that is transmitted to the feature management platform. Subsequently, the predication and stateful feature can be transmitted to other computing devices.

Abstract: Certain aspects of the present disclosure provide techniques for operation of a feature management platform. A feature management platform is an end-to-end platform developed to manage the full lifecycle of data features. For example, to create a stateful feature, the feature management platform can receive a processing artifact from a computing device. The processing artifact defines the stateful feature, including the data source to retrieve event data from, when to retrieve the event data, the type of transform to apply, etc. Based on the processing artifact, the feature management system generates a processing job (e.g., the API defines a pipeline), which when initiated generates a vector that encapsulates the stateful feature. The vector is transmitted to the computing device that locally hosts a model, which generates a prediction that is transmitted to the feature management platform. Subsequently, the predication and stateful feature can be transmitted to other computing devices.

Abstract: Certain aspects of the present disclosure provide techniques for operation of a feature management platform. A feature management platform is an end-to-end platform developed to manage the full lifecycle of data features. For example, to create a stateful feature, the feature management platform can receive a processing artifact from a computing device. The processing artifact defines the stateful feature, including the data source to retrieve event data from, when to retrieve the event data, the type of transform to apply, etc. Based on the processing artifact, the feature management system generates a processing job (e.g., the API defines a pipeline), which when initiated generates a vector that encapsulates the stateful feature. The vector is transmitted to the computing device that locally hosts a model, which generates a prediction that is transmitted to the feature management platform. Subsequently, the predication and stateful feature can be transmitted to other computing devices.

Abstract: A method including extracting, from an input, supported data. The input includes outputs from machine learning models in different formats. The supported data includes a subset of the input after data normalization. The method also includes inferring, from the supported data, data types to be used with respect to generating metrics for the machine learning models. The method also includes generating, from the supported data and using the data types, a relational event including the supported data. The relational event further includes a first data structure object including the types and having a first data structure different than the different formats. The method also includes calculating, using the supported data in the first data structure, the metrics for the machine learning models. The method also includes generating, from the relational event, a monitoring event. The monitoring event includes a second data structure object segmented into data buckets storing the metrics.

Abstract: Techniques for automatically linking pages in a web site are provided. In one technique, training data for a machine-learned scoring model is generated that comprises a plurality of features related to content items. The training data comprises multiple entries, each corresponding to a different content item in a first set of content items. For each entry, a corresponding label is based on a ranking of the corresponding content item in one or more search engine results. The machine-learned scoring model is trained based on the training data. For each content item in a second set of content items, multiple attribute values associated with that content item are input into the machine-learned scoring model, which generates a result. Based on multiple results, determining, for a particular web page, a strict subset of the second set of content items to which the particular web page will include one or more links.

Abstract: Techniques for automatically linking pages in a web site are provided. In one technique, training data for a machine-learned scoring model is generated that comprises a plurality of features related to content items. The training data comprises multiple entries, each corresponding to a different content item in a first set of content items. For each entry, a corresponding label is based on a ranking of the corresponding content item in one or more search engine results. The machine-learned scoring model is trained based on the training data. For each content item in a second set of content items, multiple attribute values associated with that content item are input into the machine-learned scoring model, which generates a result. Based on multiple results, determining, for a particular web page, a strict subset of the second set of content items to which the particular web page will include one or more links.

Abstract: Techniques are provided for determining predicted results for entities based on relatedness of the entities in a graph of nodes. In an embodiment, the graph of nodes is generated based on the determined relatedness of the entities. A node in the graph of nodes represents an entity, and nodes representing entities with known results are assigned those results as their respective node values. The assigned node values are then propagated between the neighboring nodes throughout the graph of nodes in the amount determined by the relatedness of the nodes. Based on the propagation, node values for entities with unknown results are determined and represent the predicted results for those entities. Additionally, various classifiers may be combined with the propagated node values to increase the accuracy of the predicted results.

Abstract: In an example embodiment, a supervised machine learning algorithm is used to train a communication reply score model based on an extracted first set of features and second set of features from social networking service member profiles and activity and usage information. When a plurality of member search results is to be displayed, for the member identified in each of the plurality of member search results, the member profile corresponding to the member is parsed to extract a third set of one or more features from the member profile, activity and usage information pertaining to actions taken by the members on the social networking service is parsed to extract a fourth set of one or more features, and the extracted third set of features and fourth set of features is inputted into the communication reply score model to generate a communication reply score, which is displayed visually to a searcher.

Abstract: Techniques are provided for determining predicted results for entities based on relatedness of the entities in a graph of nodes. In an embodiment, the graph of nodes is generated based on the determined relatedness of the entities. A node in the graph of nodes represents an entity, and nodes representing entities with known results are assigned those results as their respective node values. The assigned node values are then propagated between the neighboring nodes throughout the graph of nodes in the amount determined by the relatedness of the nodes. Based on the propagation, node values for entities with unknown results are determined and represent the predicted results for those entities. Additionally, various classifiers may be combined with the propagated node values to increase the accuracy of the predicted results.

Abstract: In an example embodiment, a supervised machine learning algorithm is used to train a communication reply score model based on an extracted first set of features and second set of features from social networking service member profiles and activity and usage information. When a plurality of member search results is to be displayed, for the member identified in each of the plurality of member search results, the member profile corresponding to the member is parsed to extract a third set of one or more features from the member profile, activity and usage information pertaining to actions taken by the members on the social networking service is parsed to extract a fourth set of one or more features, and the extracted third set of features and fourth set of features is inputted into the communication reply score model to generate a communication reply score, which is displayed visually to a searcher.