Salesforce, Inc.

Microsoft Technology Licensing, LLC

Yahoo! Inc.

Kin Fai Kan

Shubha Umesh Nabar

Mayukh Bhaowal

Vitaly Gordon

Leah McGuire

Till Christian Bergmann

Michael Weil

Xiaodan Sun

Christopher Rupley

Uri Merhav

Craig Martell

Gloria Lau

Bing Zhao

Peter Nicholas Skomoroch

Sarah Aerni

Vijay Krishna Narayanan

Zhigang Hua

Sara Beth Asher

Andrew Edward Kae

Saveliy Uryasev

Chaney Lin

James Reber Lewis

Nihar Dandekar

Chalenge Masekera

John Emery Ball

Subha Nabar

Luke Sedney

Arpit Amar Goel

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

The system and methods of the disclosed subject matter provide an experimentation framework to allow a user to perform machine learning experiments on tenant data within a multi-tenant database system. The system may provide an experimental interface to allow modification of machine learning algorithms, machine learning parameters, and tenant data fields. The user may be prohibited from viewing any of the tenant data or may be permitted to view only a portion of the tenant data. Upon generating an experimental model using the experimental interface, the user may view results comparing the performance of the experimental model with a current production model.

Framework for multi-tenant data science experiments at-scale

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

Techniques for inferring the identity (e.g., member profile attributes) of members of an online social network service are described. According to various embodiments, a member profile attribute missing from a member profile page associated with a particular member of an online social network service is identified. Member profile data and behavioral log data associated with a plurality of members of the online social network service is then accessed. Thereafter, a prediction modeling process is performed, based on a prediction model and feature data including the member profile data and the behavioral log data, to generate a confidence score associated with the particular member and the missing member profile attribute, the confidence score indicating a likelihood that the missing member profile attribute corresponds to a candidate value.

Inferred identity

Techniques for identifying multilingual cognates and using the multilingual cognates are provided. In one technique, multilingual cognates identified from multiple user profiles are used to train one or more translation models. In another technique, multilingual cognates identified from a single user's profile are used to translate text provided by that user. In another technique, multilingual cognates from a single user are used to align sentences in one language to sentences in another language and the aligned sentences are used to train a language model. In another technique, multilingual cognates identified from multiple user profiles are used to expand search queries. In another technique, multilingual cognates identified from multiple user profiles are used to translate other users' profiles into a target language so that users associated with a source language are viewing the other users' profiles.

Data mining multilingual and contextual cognates from user profiles

A title standardization system is may be configured to detect an edit operation associated with the job title field of a member profile stored by an on-line social network system and, in response, perform operations to derive a canonical title that represents a raw title string found in the job title field. The derived canonical title may be then associated with the member profile, in which the originally-obtained subject title string was found. This association may be stored in a database for future use, e.g., for targeting job recommendations, recruiting, making professional contacts, as well as for other purposes.

Title standardization

A system and method for automatically classifying ads into a taxonomy of categories, the method including: extracting text features from ad images using OCR (optical character recognition) techniques; identifying objects of interest from ad images using object detection and recognition techniques in computer vision; extracting text features from the web-page of the advertiser to which the user is re-directed when clicking the ad; training statistical models using the extracted features mentioned above as well as advertiser attributes from a historical dataset of ads labeled by human editors; and determining the relevant categories of unlabeled ads using the trained models.

Automatic classification of display ads using ad images and landing pages

Growing community of inventors

Sunnyvale, CA, United States of America

Kin Fai Kan