Microsoft Technology Licensing, LLC

Salesforce, Inc.

Salesforce Com, Inc.

Shubha Umesh Nabar

Michael Ching

Vitaly Gordon

Kin Fai Kan

Leah McGuire

Rahul Nair

Simon P King

Sandy Wong

Sarah Aerni

Jun Yin

Xavier Legros

Kevin L Haas

Mayukh Bhaowal

Srinivasan Badrinarayanan

Wei Mu

Sara Beth Asher

Daniel Tunkelang

Peter Bailey

Scott Joseph Counts

Nikhil Dandekar

Chad Carson

Ho John Lee

Kevin Moore

Brian Brechbuhl

Till Christian Bergmann

Michael Weil

Paul Alexander Dow

Hyun-Ju Seo

Xiaodan Sun

Christopher Rupley

Chaney Lin

Shuang Guo

Jian Qiao

Leslie Fine

Aditya Pal

Xiuchai (Meko) Xu

Eric Wayman

James Reber Lewis

Nihar Dandekar

Chalenge Masekera

John Emery Ball

Rick Munoz

John Grotland

Melissa Runfeldt

Natalie Casey

Seiji J Yamamoto

A set of data for training a machine learning system can be modified to improve its performance. An item of information can be transmitted. A message can be transmitted that includes an explanation of a determination, by the machine learning system, to transmit the item of information from among a plurality of items of information. A first set of data can have been used to train the machine learning system. A signal can be received that includes an indication of a usefulness of the message, to a user of a user device, in making a decision to perform an action based on a knowledge associated with the item of information. The first set of data can be modified, in response to a receipt of the signal, to produce a second set of data. The machine learning system can be caused to be trained using the second set of data.

Automatic generation of an explanation, based on data lineage and user feedback, of a prediction produced by a machine learning system

A method for generating a model for recommendations from an item data set for a target data set includes embedding a set of targets from the target data set in a shared coordinate space using a first embedding function, embedding a first set of items from the item data set in the shared coordinate space using a second embedding function, selecting at least one target from the set of targets, and identifying a second set of items from the first set of items that are proximate to the at least one target as candidates from the recommendations.

Multi-tenant, metadata-driven recommendation system

A system may automatically generate a predictive machine learning model by automatically performing various processes based on an analysis of the data as well as metadata associated with the data. The system may accept a selection of data and a prediction field from the data. The system may automatically generate a set of features based on the data and may automatically remove certain features that cause inaccuracies in the model. The system may balance the data based on a representation rate of certain outcomes. The system may train and select a model based on several candidate models. The system may then perform the predictions based on the selected model and send an indication of the predictions to a user.

Automatic machine learning model generation

Methods and systems are provided to determine suitable hyperparameters for a machine learning model and/or feature engineering process. A suitable machine learning model and associated hyperparameters are determined by analyzing a dataset. Suitable hyperparameter values for compatible machine learning models having one or more hyperparameters in common and a compatible dataset schema are identified. Hyperparameters may be ranked according to each of their respective influences on a model performance metrics, and hyperparameter values identified as having greater influence may be more aggressively searched.

Identification and application of hyperparameters for machine learning

A method of training a predictive model to predict a likely field value for one or more user selected fields within an application. The method comprises providing a user interface for user selection of the one or more user selected fields within the application; analyzing a pre-existing, user provided data set of objects; training, based on the analysis, the predictive model; determining, for each user selected field based on the analysis, a confidence function for the predictive model that identifies the percentage of cases predicted correctly at different applied confidence levels, the percentage of cases predicted incorrectly at different applied confidence levels, and the percentage of cases in which the prediction model could not provide a prediction at different applied confidence levels; and providing a user interface for user review of the confidence functions for user selection of confidence threshold levels to be used with the predictive model.

Recognition of biases in data and models

A method for improving mass messaging in an electronic messaging system includes receiving recipient data describing a response of each of one or more recipients to receiving a prior message, generating predictor data based on the recipient data, where the predictor data indicates a plurality of predictors of recipient behavior in response to a message, identifying one or more top predictors of recipient behavior, the one or more top predictors being selected from among the plurality of predictors based on preferred recipient behaviors, generating, for each of the one or more recipients and from the recipient data, one or more predictive scores for each combination of top predictor and recipient, and assigning, based on one or more predictive scores of a specific recipient, the specific recipient to a specific persona, wherein the specific persona describes an expected behavior of the recipient.

Predictive scoring and messaging in messaging systems

Techniques for determining an editable query which, when executed, identifies and presents a set of member profiles similar to a source member profile are described. In an embodiment, a preliminary set of member profiles similar to a source member profile is identified. A set of profile features is obtained from the preliminary set of similar member profiles, and a set of relevant profile features is extracted from the set of profile features. An editable query including the set of relevant profile features is presented. In an embodiment, the set of member profiles similar to the source member profile and corresponding to the set of relevant profile features is also presented. Upon having been edited, the editable query is executable to identify and present a new set of member profiles corresponding to the edited query.

Query-by-example for finding similar people

Search results may include both objective results and person results. In one example, a search query is evaluated to determine whether it is the type of query that a user might want to ask to a friend. If the query is of such a type, then the search engine may examine a social graph to determine which friends of the user who entered the query may have information that is relevant to answering the query. If such friends exist, then the friends may be displayed along with objective search results, along with an explanation of each friend's relevance to the query. Clicking on a person in the results may cause a conversation to be initiated with that person, thereby allowing the user who entered the query to ask his or her friend about the subject of the query.

Providing objective and people results for search

The author ranking technique described herein is a technique to rank authors in social media systems along various dimensions, using a variety of statistical methods for utilizing those dimensions. More particularly, the technique ranks authors in social media systems through a combination of statistical techniques that leverage usage metrics, and social and topical graph characteristics. In various exemplary embodiments, the technique can rank author authority by the following: 1) temporal analysis of link sharing in which authority is computed based on a user's propensity to provide early links to web pages that subsequently become popular; 2) topical authority based on the author's links and content updates in specific topic areas; and 3) popularity and influence based on nodal properties of authors.

Ranking authors in social media systems

To determine a configuration for a database system, a plurality of queries may be sampled from a representative workload using statistical inference to compute the probability of correctly selecting one of a plurality of evaluation configurations. The probability of correctly selecting may determine which and/or how many queries to sample, and/or may be compared to a target probability threshold to determine if more queries must be sampled. The configuration from the plurality of configurations with the lowest estimated cost of executing the representative workload may be determined based on the probability of selecting correctly. Estimator variance may be reduced through a stratified sampling scheme that leverages commonality, such as an average cost of execution, between queries based on query templates. The applicability of the Central Limit Theorem may be verified and used to determine which and/or how many queries to sample.

Database configuration analysis

Growing community of inventors

Mountain View, CA, United States of America

Shubha Umesh Nabar