Adobe, Inc.

Charles Menguy

Anup Bandigadi Rao

Kourosh Modarresi

Virgil-Artimon Palanciuc

Margarita Savova

Hongyuan Yuan

Ryan A Rossi

Fan Du

Catalin Costache

Yeuk-Yin Chan

Mihai Daniel Fecioru

Eunyee Koh

Tung Mai

Matvey Kapilevich

David Meier Weinstein

Moumita Sinha

Ravi Prakash Singh

Vijeth Lomada

Harleen S Sahni

Edward Schuchardt

Lakshmi Shivalingaiah

Sudhakar Pandey

Roshan Singh

Navin Chaganti

Ankush Sharma

Aditya Satishrao Borde

Liron Goren Snai

Multi-modal machine-learning model training techniques for search are described that overcome conventional challenges and inefficiencies to support real time output, which is not possible in conventional training techniques. In one example, a search system is configured to support multi-modal machine-learning model training. This includes use of a preview mode and an expanded mode. In the preview mode, a preview segment is generated as part of real time training of a machine learning model. In the expanded mode, the preview segment is persisted as an expanded segment that is used to train and utilize an expanded machine-learning model as part of search.

Multi-modal machine-learning model training for search

A cluster generation system identifies data elements, from a first binary record, that each have a particular value and correspond to respective binary traits. A candidate description function describing the binary traits is generated, the candidate description function including a model factor that describes the data elements. Responsive to determining that a second record has additional data elements having the particular value and corresponding to the respective binary traits, the candidate description function is modified to indicate that the model factor describes the additional elements. The candidate description function is also modified to include a correction factor describing an additional binary trait excluded from the respective binary traits. Based on the modified candidate description function, the cluster generation system generates a data summary cluster, which includes a compact representation of the binary traits of the data elements and additional data elements.

Trait expansion techniques in binary matrix datasets

Systems, methods, and non-transitory computer-readable media (systems) are disclosed for generating meaningful and insightful user segment reports based on a high dimensional data space. In particular, in one or more embodiments, the disclosed systems utilize a relaxed bi-clustering model to automatically identify user segments in a data space including datasets of features specific to individual users. In at least one embodiment, the disclosed systems identify and include users in automatically generated user segments even though those users are associated with some, but perhaps not all, of the features as other members in the automatically generated user segments.

Automatically generating user segments

Techniques disclosed herein relate generally to evaluating and selecting candidate datasets for use by software applications, such as selecting candidate datasets for training machine-learning models used in software applications. Various machine-learning and other data science techniques are used to identify unique entities in a candidate dataset that are likely to be part of target entities for a software application. A merit attribute is then determined for the candidate dataset based on the number of unique entities that are likely to be part of the target entities, and weights associated with these unique entities. The merit attribute is used to identify the most efficient or most cost-effective candidate dataset for the software application.

Machine-learning techniques for evaluating suitability of candidate datasets for target applications

The present disclosure describes systems, non-transitory computer-readable media, and methods for utilizing hash partitions to determine local densities and distances among users (or among other represented data points) for clustering sparse data into segments. For instance, the disclosed systems can generate hash signatures for users in a sparse dataset and can map users to hash partitions based on the hash signatures. The disclosed systems can further determine local densities and separation distances for particular users (or other represented data points) within the hash partitions. Upon determining local densities and separation distances for datapoints from the dataset, the disclosed systems can select a segment (or cluster of data points) grouped according to a hierarchy of a clustering algorithm, such as a density-peaks-clustering algorithm.

Segmenting users with sparse data utilizing hash partitions

Certain embodiments involve generating a real-time notification to facilitate the delivery of customized content, based on detecting that a subject activity occurs within a customized subject region. For instance, a computing system updates a graphical interface to display, as a layer on a map, detected instances of a subject activity performed within a geographic area of the map. The computing system determines a polygon corresponding to a region of the geographic area, where the polygon is determined from a graphical input representing lines at locations on the map on which the detected instances are overlaid. The computing system determines that a location of a detected instance of the subject activity performed by a user device falls within the polygon. The computing system transmits a notification to a content provider, such that the content provider delivers customized content to the user device that performed the detected instance of the subject activity.

Customized geospatial population segmentation based on a received polygon definition

Dynamic clustering of sparse data utilizing hash partitions

Automatically generating meaningful user segments

Systems and techniques for delivery of contextual interest from interaction information are described that process user interactions with digital content to generate user interest scores for various topics. A contextual user interest system uses user interaction data to identify and contextualize content, and assigns propensity scores to the contextualized content. By dynamically contextualizing pages of content, the contextual user interest system may adapt to changes in the content and provide more accurate and robust information over time, which is not possible using conventional techniques. The contextualized pages of content are used to assign user interest scores across a number of topics to users who have visited the pages of content, and the user interest scores are normalized in a manner that allows a user's degree of interest in a topic to be compared to that of another user.

Delivery of contextual interest from interaction information

Techniques are disclosed that provide more accurate clustering of devices by forming clusters of devices and merging or changing clusters based on predetermined criteria. The technique starts with a large number of clusters (e.g., one for each account) and refines the clusters, for example, by merging clusters or determining which cluster a given device should be in when the device is associated with multiple clusters. One technique iteratively adjusts clusters of devices by merging clusters determined to be associated with a single user until a cluster contains all of the devices and accounts expected to be associated with a single user.

Identifying multiple devices belonging to a single user

Persistent profile identifiers can be produced to identify clusters of devices accessing a network in different time periods. In one embodiment, an apparatus uses a first identifier from a first group of identifiers to identify a first cluster of devices and uses a second identifier from a second group of identifiers to identify a second cluster of devices. Further, the apparatus determines that the first cluster of devices identified by the first identifier and the second cluster of devices identified by the second identifier form an edge in a maximum cluster matching. The apparatus provides the first identifier as a persistent identifier for the first cluster of devices and the second cluster of devices.

Generating persistent profile identifiers

Systems and methods are provided for identifying stable attributes in clusters of devices and naming the clusters of devices using the stable attributes. The stable attribute is selected from data associated with user/household behavior and data associated with each device. Because the stable attribute is associated with the device in the cluster determined to be the most stable, it is a persistent identifier that can be used to identify the user/household over time. Because the stable attribute is determined each time the clusters are determined, resources that would be required to track the user/household over time are saved and the system of tracking users/households is more efficient. Attributes evaluated to determine the stable attribute include IP address, the location of the device, hardware characteristics of the device, the date a CRM profile was created, timestamp of the first or last user authentication, and the number of authentications on that device.

Techniques for associating variable versions of a cluster of devices identified at different times with the same user by assigning cluster identifiers

Growing community of inventors

New York, NY, United States of America

Charles Menguy