Abstract: An electronic system stores metadata for a plurality of media items, including, for each media item of the plurality of media items, at least one categorical identifier from a set of categorical identifiers. For a user of the media-providing service, the electronic system (i) determines a distribution of interests of the user with respect to the set of categorical identifiers; (ii) generates a network graph configured to represent a calibrated media item selection task, wherein the network graph represents respective relevance scores for each respective media item of the plurality of media items and the distribution of interests of the user with respect to the categorical identifiers; (iii) selects a set of media items from the plurality of media items to recommend to the user by solving for a maximum flow of the network graph; and (iv) provides the set of media items as recommendations to the user.

Abstract: 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.

Abstract: 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.

Abstract: Methods, systems, and devices for data attribution using frequent pattern analysis are described. In some cases, data stored at a multi-tenant database server may be analyzed to understand various interactions and patterns between data attributes associated with multiple users. The multi-tenant database server may effectively cluster and/or perform calculations on attributes of the data to understand user patterns. In some examples, the multi-tenant database server may determine a change (e.g., a probability change) in the user patterns by removing one or more attributes from the data set and re-performing the analysis. By re-performing the analysis, the multi-tenant database server may attribute a value to individual pieces and combinations of the data in order to indicate the effect that each piece of data has on the analysis.

Abstract: 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.

Abstract: Methods, systems, and devices for data attribution using frequent pattern analysis are described. In some cases, data stored at a multi-tenant database server may be analyzed to understand various interactions and patterns between data attributes associated with multiple users. The multi-tenant database server may effectively cluster and/or perform calculations on attributes of the data to understand user patterns. In some examples, the multi-tenant database server may determine a change (e.g., a probability change) in the user patterns by removing one or more attributes from the data set and re-performing the analysis. By re-performing the analysis, the multi-tenant database server may attribute a value to individual pieces and combinations of the data in order to indicate the effect that each piece of data has on the analysis.