Adobe, Inc.

Ritwik Sinha

Shiv Kumar Saini

Dhruv Singal

Sunny Dhamnani

Meghanath Macha Yadagiri

Said Kobeissi

Balaji Vasan Srinivasan

Vishwa Vinay

Atanu Ranjan Sinha

John Bevil Bates

Anandhavelu Natarajan

Prakhar Gupta

Pranav Ravindra Maneriker

Anandhavelu N

Vineet Gupta

Shriram Revankar

Lilly Kumari

Sourav Pal

Ajaykrishnan Jayagopal

Shubh Gupta

Ryan A Rossi

Wei Zhang

William Brandon George

Anup Bandigadi Rao

Fan Du

Deepak Pai

Tung Mai

Iftikhar Ahamath Burhanuddin

Trevor Paulsen

Sunav Choudhary

Kushal Chawla

Matvey Kapilevich

David Meier Weinstein

David Arbour

Sanket Vaibhav Mehta

Margarita Savova

Virgil-Artimon Palanciuc

Yasin Abbasi Yadkori

Ivan Andrus

Tharun Mohandoss

Scott Allen Tomko

Balaraman Ravindran

Chris Challis

Sayaji Hande

Michael Rimer

Prithvi Bhutani

Alexandru Ionut Hodorogea

Yash Shrivastava

Aahlad Manas Puli

Mike Rimer

Manish Dash

Michael Young

Piyush Kumar

Amerineni Rohith

Rajat Kateja

Sandeep Zechariah George

Harvineet Singh

Determination of high value customer journey sequences is performed by determining customer interactions that are most frequent as length N=1 sub-sequences, recursively determining most frequent length N+1 sub-sequences that start with the length N sub-sequences, determining a first count indicating how often one of the sub-sequences appears in the sequences, determining a second count indicating how often the one sub-sequence resulted in the goal, and using the counts to determine the most or least effective sub-sequences for achieving the goal.

Intelligent customer journey mining and mapping

An IDS generator determines multiple classes for electronic data items. The IDS generator determines, for each class, a class-specific candidate ruleset. The IDS generator performs a differential analysis of each class-specific candidate ruleset. The differential analysis is based on differences between result values of a scoring objective function. In some cases, the differential analysis determines at least one of the differences based on additional data structures, such as an augmented frequent-pattern tree. A probability function based on the differences is compared to a threshold probability At least one testing ruleset is modified based on the comparison. The IDS generator determines, for each class, a class-specific optimized ruleset based on the differential analysis of each class-specific candidate ruleset. The IDS generator creates an optimized interpretable decision set based on combined class-specific optimized rulesets for the multiple classes.

Fast and accurate rule selection for interpretable decision sets

A content saliency network is a machine-learned neural network that predicts the saliency of elements of a content item. The content saliency network may be used in a method that includes determining a set of elements in draft content and computing a first pixel-level vector for the content. The method may also include, for each element in the set of elements, computing a vector of simple features for the element, the simple features being computed from attributes of the element, computing a second pixel-level vector for the element, computing a third pixel-level vector for an intermediate context of the element, and providing the vectors to the content saliency network. The content saliency network provides a saliency score for the element. The method further includes generating an element-level saliency map of the content using the respective saliency scores for the set of elements and providing the saliency map to a requestor.

Saliency prediction for informational documents

Systems and methods are described for a causal marketing attribution process that includes the receiving of a plurality of marketing events associated with a customer and computing a sum of a plurality of channel-specific terms corresponding to the plurality of marketing events, wherein each of the plurality of channel-specific terms comprises a channel-specific base parameter and a channel-specific decay parameter. Additionally, the causal marketing attribution process computes a sum of a plurality of interaction terms, wherein each interaction term comprises a product of a pair of channel-specific terms, and determines a probability of a target outcome for the customer based on the sum of the plurality of channel-specific terms and the sum of the plurality of interaction terms.

Marketing attribution capturing synergistic effects between channels

Various embodiments describe user segmentation. In an example, potential rules are generated by applying a frequency-based analysis to user interaction data points. Each of the potential rules includes a set of attributes of the user interaction data points and indicates that these data points belong to a segment of interest. An objective function is used to select an optimal set of rules from the potential rules for the segment of interest. The potential rules are used as variable inputs to the objective function and this function is optimized based on interpretability and accuracy parameters. Each rule from the optimal set is associated with a group of the segment of interest. The user interaction data points are segments into the groups by matching attributes of these data points with the rules.

Accurate and interpretable rules for user segmentation

This disclosure involves allocating content-delivery resources to electronic content-delivery channels based on attribution models accuracy. For instance, a simulation is executed that involves simulating user exposures, times between user exposures, and user responses. The simulation is performed based on parameters associated with simulating user exposures to electronic content-delivery channels and user responses to the user exposures. An accuracy of a channel attribution model when estimating an attribution of an electronic content-delivery channel to a user response is evaluated based on the simulation. A channel attribution model is selected based on the evaluation. An attribution of the electronic content-delivery channel is determined by applying the selected channel attribution model to actual user exposures and actual user responses. This attribution can be used to allocate content-delivery resources to the electronic content-delivery channel in accordance with the selected channel attribution model, and thereby provide interactive content via the electronic content-delivery channel.

Simulation-based evaluation of a marketing channel attribution model

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

Forecasting a potential audience size and an unduplicated audience size for a digital campaign includes receiving an audience segment input and a time period input. The audience segment input is converted into multiple atomic target specifications. For each of the multiple atomic target specifications, a potential audience size is determined during the time period input by selecting a time series model based on a frequency of attribute values from the atomic target specification and combining the selected time series model with a frequent item set model. The potential audience size for each of atomic target specifications is aggregated over the time period input into a total potential audience size. The total potential audience size is output. The time series model and the frequent item set model are obtained using data from a historic bid request database.

Forecasting potential audience size and unduplicated audience size

Systems and techniques for identifying segments in categorical data include receiving multiple transaction ID (TID) lists with univariate values that satisfy a thresholding metric with each TID list representing an occurrence of a single attribute in a set of transactions. The TID lists are stored with the univariate values that satisfy the thresholding metric in a data structure. In a loop, candidate itemsets to form from combinations of TID lists are determined using only the combinations of TID lists that satisfy categorical constraints. In the loop, for the candidate itemsets that satisfy categorical constraints, both the thresholding metric and a similarity metric are applied to the candidate itemsets. Final itemsets are formed from only the candidate itemsets that satisfy both the thresholding metric and the similarity metric.

Identifying high value segments in categorical data

Techniques are disclosed for identifying, assessing, and presenting differences between segments of customers. The techniques identify differences in categorical features of the customers in two segments. The techniques use observed data to identify differences in a categorical feature. The techniques then assess whether the observed difference is a real difference applicable to the entire customer population or the result of random chance. The categorical features with the most significant differences (i.e., unlikely due to random chance) are presented, for example, to allow a marketer to easily appreciate the most significant segment differences. Certain techniques account for segment overlap (i.e., customers being in both segments) in assessing whether differences are due to random chance. Certain techniques limit the presented categorical features to account for common knowledge and/or false testing issues. Certain techniques present segment differences incrementally during the computations to provide quicker access to relevant information.

Detecting differing categorical features when comparing segments

Methods, systems, and non-transitory computer readable storage media are disclosed for detecting robotic activity while monitoring Internet traffic across a plurality of domains. For example, the disclosed system identifies network session data for each domain of a plurality of domains, the network session data including network sessions comprising features that indicate human activity. In one or more embodiments, the disclosed system generates a classifier to output a probability that a network session at a domain includes human activity. In one or more embodiments, the disclosed system also generates a classifier to output a probability that a network session includes good robotic activity. Additionally, the disclosed system generates a domain-agnostic machine-learning model by combining models from a plurality of domains with network sessions including human activity.

Detecting robotic internet activity across domains utilizing one-class and domain adaptation machine-learning models

A session identification system classifies network sessions with a network application as either human-generated or generated by a non-human, such as by a bot. In an embodiment, the session identification system receives a set of unlabeled network sessions, and determines a label for a single class of the unlabeled network sessions. Based on the one-class labeling information, the session identification system determines multiple subsets of the unlabeled network sessions. Multiple classifiers included in the session identification system generate probabilities describing each of the unlabeled network sessions. The session identification system classifies each of the unlabeled network sessions based on a combination of the generated probabilities.

Classification of website sessions using one-class labeling techniques

A content saliency network is a machine-learned neural network that predicts the saliency of elements of a content item. The content saliency network may be used in a method that includes determining a set of elements in a UI and computing a first context vector for the content. The method may also include, for each element in the set of elements, computing a vector of simple features for the element, the simple features being computed from attributes of the element, computing a second context vector for the element, computing a third context vector for an intermediate context of the element, and providing the vectors to the content saliency network. The content saliency network provides a saliency score for the element. The method further includes generating an element-level saliency map of the content using the respective saliency scores for the set of elements and providing the saliency map to a requestor.

Saliency prediction for a mobile user interface

Systems and methods disclosed herein identify multivariate relationships that exist across all types data collected from numerous observed users over one or more networks. Electronic data collected from observed users include categorical data and non-categorical/numeric data. To compare and analyze the collected data, a marketing entity converts the numeric data to categorical data via a binning algorithm, which reduces the numeric data into two or more discrete categories. The marketing entity analyzes the data variables to compute pairwise associations on the collected categorical and numeric data (which has been converted to categorical data). The marketing entity also determines hierarchical clusters to group the pairwise associations of data variables based on the strength of the associations. The pairwise relationships and hierarchical clusters are displayed on a user interface.

Systems and techniques for determining associations between multiple types of data in large data sets

The present disclosure is directed toward systems and methods for identifying contributing factors associated with a multi-variable metric anomaly. One or more embodiments described herein identify one or more contributing factors that led to an anomaly in a multi-variable metric by calculating linearizing weights such that the total deviation in the multi-variable metric can be written as a weighted sum of deviations for dimension elements associated with the multi-variable metric.

Identifying sources of anomalies in multi-variable metrics using linearization

Growing community of inventors

Kolkata, India

Ritwik Sinha