Abstract: Training and/or utilizing a machine learning model to generate request agnostic predicted interaction scores for electronic communications, and to utilization of request agnostic predicted interaction scores in determining whether, and/or how, to provide corresponding electronic communications to a client device in response to a request. A request agnostic predicted interaction score for an electronic communication provides an indication of quality of the communication, and is generated independent of corresponding request(s) for which it is utilized. In many implementations, a request agnostic predicted interaction score for an electronic communication is generated “offline” relative to corresponding request(s) for which it is utilized, and is pre-indexed with (or otherwise assigned to) the electronic communication. This enables fast and efficient retrieval, and utilization, of the request agnostic interaction score by computing device(s), when the electronic communication is responsive to a request.

Abstract: An optimized and efficient method of identifying one or more points within a dataset that are close to the centers of clumps similar records in a large, multi-element dataset uses Monte Carlo techniques to compute approximate clustering costs at significantly reduced computational expense. The inaccuracy caused by the approximate methods is also estimated, and if it is too high, the method may be repeated with a larger Monte Carlo sample size to improve accuracy. Portions of the algorithm that are independent are distributed among a number of cooperating computing nodes so that the full algorithm can be completed in less time.

Abstract: An optimized and efficient method of identifying one or more points within a dataset that are close to the centers of clumps similar records in a large, multi-element dataset uses Monte Carlo techniques to compute approximate clustering costs at significantly reduced computational expense. The inaccuracy caused by the approximate methods is also estimated, and if it is too high, the method may be repeated with a larger Monte Carlo sample size to improve accuracy.

Abstract: Methods, systems, apparatus, including computer programs encoded on computer storage medium, for a bottom-up approach for generating high-quality content streams. In one aspect, the method includes actions of obtaining data identifying a plurality of content items, generating a plurality of queries for the particular topic, and for each query of the plurality of queries: obtaining a set of search results for the query that identify content items identified in the obtained data, and determining, from the search results for the query, a respective quality score for each of one or more quality characteristics. The method may also include actions such as identifying one or more first high-quality queries from the plurality of queries based on the respective quality scores for the one or more quality characteristics, and populating a stream of content for display on the user device using search results for the one or more first high-quality queries.

Abstract: An optimized and efficient method of identifying one or more points within a dataset that are close to the centers of clumps similar records in a large, multi-element dataset uses Monte Carlo techniques to compute approximate clustering costs at significantly reduced computational expense. The inaccuracy caused by the approximate methods is also estimated, and if it is too high, the method may be repeated with a larger Monte Carlo sample size to improve accuracy. Portions of the algorithm that are independent are distributed among a number of cooperating computing nodes so that the full algorithm can be completed in less time.

Abstract: Training and/or utilizing a machine learning model to generate request agnostic predicted interaction scores for electronic communications, and to utilization of request agnostic predicted interaction scores in determining whether, and/or how, to provide corresponding electronic communications to a client device in response to a request. A request agnostic predicted interaction score for an electronic communication provides an indication of quality of the communication, and is generated independent of corresponding request(s) for which it is utilized. In many implementations, a request agnostic predicted interaction score for an electronic communication is generated “offline” relative to corresponding request(s) for which it is utilized, and is pre-indexed with (or otherwise assigned to) the electronic communication. This enables fast and efficient retrieval, and utilization, of the request agnostic interaction score by computing device(s), when the electronic communication is responsive to a request.

Abstract: An optimized and efficient method of identifying one or more points within a dataset that are close to the centers of clumps similar records in a large, multi-element dataset uses Monte Carlo techniques to compute approximate clustering costs at significantly reduced computational expense. The inaccuracy caused by the approximate methods is also estimated, and if it is too high, the method may be repeated with a larger Monte Carlo sample size to improve accuracy.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for feature-based ranking adjustment. In one aspect, a method includes finalizing rankings of resources based on detected features, and for each resource for which a ranking is not finalized, finalizing the respective resources or demoting the resources based on the detection of features common to the resources with the finalized rankings and the resources with the unfinalized rankings.

Abstract: Embodiments include identifying a plurality of communication sessions, each of the plurality of communication sessions having multiple participants. Embodiments may also include applying at least one parameter to each of the communication sessions and determining a non-adjusted value for the at least one parameter for each of the communication sessions. Embodiments may further include determining whether the at least one parameter has been assigned a predetermined weight adjustment. If the at least one parameter has been assigned a weight adjustment, adjusting the value for the at least one parameter based upon the predetermined weight assigned to each of the at least one parameter. Embodiments may include generating a combined value for each of the communication sessions based upon, at least in part, one or more of the non-adjusted value and the adjusted value and ranking each of the communication sessions based upon, at least in part, the combined value.