Adobe, Inc.

Matvey Kapilevich

Anup Bandigadi Rao

Tung Mai

Margarita Savova

Ryan A Rossi

Nedim Lipka

Charles Menguy

David Meier Weinstein

Moumita Sinha

Karthik Raman

Harleen S Sahni

Fan Du

Ritwik Sinha

Yasin Abbasi Yadkori

David Allen Jared

Yeuk-Yin Chan

Vijeth Lomada

Lakshmi Shivalingaiah

Alexandru Ionut Hodorogea

Nicholas M Jordan

Nicholas M Jordon

Paul Mackles

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

The present disclosure relates to systems, methods, and non-transitory computer-readable media that estimate the overlap between sets of data samples. In particular, in one or more embodiments, the disclosed systems utilize a sketch-based sampling routine and a flexible, accurate estimator to determine the overlap (e.g., the intersection) between sets of data samples. For example, in some implementations, the disclosed systems generate a sketch vector&#x2014;such as a one permutation hashing vector&#x2014;for each set of data samples. The disclosed systems further compare the sketch vectors to determine an equal bin similarity estimator, a lesser bin similarity estimator, and a greater bin similarity estimator. The disclosed systems utilize one or more of the determined similarity estimators in generating an overlap estimation for the sets of data samples.

Generating overlap estimations between high-volume digital data sets based on multiple sketch vector similarity estimators

The present disclosure relates to training a recommendation model to generate trait recommendations using one permutation hashing and populated-value-slot-based densification. In particular, the disclosed systems can train the recommendation model by computing sketch vectors corresponding to traits using one permutation hashing. The disclosed systems can then fill in unpopulated value slots of the sketch vectors using populated-value-slot-based densification. The disclosed systems can combine the resulting densified sketches to generate the trained recommendation model. For example, in some embodiments, the disclosed systems can combine the sketches by generating a plurality of locality sensitive hashing tables based on the sketches. In some embodiments, the disclosed systems generate a count sketch matrix based on the sketches and generate trait embeddings based on the count sketch matrix using spectral embedding. Based on the trait embeddings, the disclosed systems can utilize the recommendation model to flexibly and accurately determine the similarity between traits.

Utilizing one hash permutation and populated-value-slot-based densification for generating audience segment trait recommendations

Systems and methods are disclosed for clustering multiple devices that are associated with particular users by utilizing both probabilistic and deterministic data derived from analytics information on the users. An analytics computing system generates at least one deterministic device cluster that groups a first set of devices associated with a first user. The first set of devices share deterministic user identifiers specific to the first user. The analytics computing system also identifies a probabilistic link between a device in the first set of devices and additional devices. The probabilistic link indicates common usage patterns between two devices. Based on the probabilistic link, the analytics computing system generates a data structure that includes the deterministic device cluster and the additional devices.

Methods and systems for identifying multiple devices belonging to a single user by merging deterministic and probabilistic data to generate a cross device data structure

Selection of a trait may be received. A complex segment rule may be created that is usable to evaluate one or more qualification events. For example, the segment rule may be usable to evaluate a combined recency and frequency of the one or more qualification events. The qualification events may be based on collected network data associated with the plurality of visitors with each qualification event corresponding to a separate qualification of visitor according to the trait. The qualification events may be evaluated together according to the segment rule. For example, the combined recency and frequency of the one or more qualification events may be evaluated according to the segment rule. Evaluating the segment rule may include estimating a segment population size in real-time.

Audience size estimation and complex segment logic

Certain embodiments involve associating a device with an identifier of a router that is assigned an IP address based on DHCP. For example, connection data that includes the IP address is analyzed to associate devices with the identifier of the router. A determination is made as to whether any usage of the IP address by the devices overlap in time. Devices with overlapping usage are determined to be connecting to online resources via a same router, while devices with non-overlapping usage are determined to be connecting from other routers. A single router identifier is associated with the devices using the same router. This identifier is then used to track the online activity of the associated devices.

Attributing online activities to internet protocol addresses of routers for customizing content to different networks

Tunable algorithmic segment techniques are described. In one or more implementations, a target audience definition is obtained that is input to initiate creation of a look-alike model. The target audience definition indicates traits associated with a baseline group of consumers who have interacted with online resources in a designated manner, such as by buying a product, visiting a website, using a service, and so forth. Tuning parameters designated for the look-alike model are ascertained and the look-alike model is built based on the target audience definition and the tuning parameters. The tuning parameters may include at least a setting selectable to control reach versus accuracy for the look-alike model. Segment data indicative of market segments generated according to the look-alike model may then be exposed for manipulation by a client. The manipulation may include selectable control over the tuning parameters to generate different look-alike groups from the segment data.

Tunable algorithmic segments

Growing community of inventors

Irvington, NY, United States of America

Matvey Kapilevich