Microsoft Technology Licensing, LLC

Shoou-Jiun Wang

Jayaram N M Nanduri

Yi Zhang

Cezary Antoni Marcjan

Jaclyn Ruth Elizabeth Phillips

Tracy My Tuyen Nguyen

Yuting Jia

Jayaram Nm Nanduri

Ram P Sunkara

Adam Feldman Reinhardt

Samuel B Anson

Stuart H Dwyer

Jennifer Olivia Ede

Weiwei Wu

Thinesh Thangakumar Abimanyu Rathinavelu

Jonathan Ray Bates

Hung-Chih Yang

Lu Xia

Sandipan Ganguly

Eslam K Abdelreheem

Yung-Wen Liu

Xing Zhang

Ming-Yu Fan

Jennifer Oliva Ede

Huiying Mao

Daniil Sokolov

Webber Po-Wei Lee

Justin Hobart

Ram Prasad Sunkara

Embodiments are provided for suggesting topics in a messaging system. A set of queries is received from a chat transcript history, where the set of queries includes a set of unhandled queries, and each unhandled query comprises a query for which a bot did not identify a corresponding topic (e.g., queries that did not trigger selection of a topic by the bot). A vector representation is generated for each unhandled query in the set of unhandled queries. The vector representations for the set of unhandled queries are clustered to generate one or more clusters of vector representations, each cluster corresponding to a group of unhandled queries. A corresponding suggested topic is generated for each cluster and provided to an authoring tool that comprises one or more interactive elements to enable an author to select at least one of the suggested topics for implementation in the bot.

Topic suggestion in messaging systems

Various device attributes associated with a current event may be obtained. Similarity metrics may be determined that indicate a degree of similarity between the device attributes that are associated with the current event and stored device attributes that are associated with previous events and previously created fuzzy device identifiers. A fuzzy device identifier may be assigned to the current event based at least in part on a comparison of the similarity metrics with a threshold. If none of the similarity metrics compare favorably with the threshold, then a new fuzzy device identifier may be created for the current event. However, if at least one of the similarity metrics compares favorably with the threshold, then the previously created fuzzy device identifier whose stored device attributes are most similar to the device attributes that are associated with the current event may be assigned to the current event.

Facilitating device fingerprinting through assignment of fuzzy device identifiers

Embodiments are provided for detecting overlapping topics in a messaging system. In an example system, a plurality of trigger phrases is received, where each trigger phrase is configured to trigger a bot that receives the trigger phrase to select a corresponding topic for conversation. For each trigger phrase, a vector representation is generated. Measures of similarity are generated based at least on the vector representations, where each measure of similarity represents a degree of similarity between a respective pair of vector representations. A topic overlap is detected based on a pair of vector representations having a measure of similarity above a similarity threshold, where the topic overlap indicates two trigger phrases that are overlapping. The topic overlap is provided to an authoring tool that comprises one or more interactive elements to enable a user to change at least one of the two trigger phrases that are overlapping.

Topic overlap detection in messaging systems

The disclosure herein describes managing defects in a model training pipeline. A synthetic data set is generated that is associated with a defect type and a lifecycle stage of the model training pipeline, and baseline performance metrics associated with the defect type are generated. Based on a code change to the pipeline, a test model is trained using the pipeline and the synthetic data set, and test performance metrics are collected based on the test model and associated with the defect type. Based on comparing the baseline performance metrics and the test performance metrics, a defect of a particular defect type is identified in the pipeline. An indicator of the defect is provided that includes the defect type and the lifecycle stage with which the synthetic data set is associated, whereby a defect correction process is enabled to remedy the defect based on the associated defect type and the lifecycle stage.

Managing defects in a model training pipeline using synthetic data sets associated with defect types

A trusted transaction signal is passed from a merchant to an account provider. A merchant receives a payment account association with a transaction between a customer and the merchant, and determines a level of risk for the transaction. If the customer is unknown to the merchant, the trust level is determined to be a default trust level. If the customer is known to the merchant, then a risk assessment determines whether the trust level is the default trust level, or a second, greater, trust level. A transaction notification is sent to the account provider. When the trust level is the default level, the notification includes a merchant ID that identifies the merchant and signifies the default trust level for the transaction. When the trust level is the greater level, the notification includes a different merchant ID that also identifies the merchant, but represents the second trust level for the transaction.

Passing a trusted transaction signal

Training risk determination models based on a set of labeled data transactions. A first set of labeled data transactions that have been labeled during a review process is accessed. A first risk determination model is trained using the first set of labeled data transactions. A first risk score for data transactions of a set of unlabeled data transactions is determined using the first risk determination model. Data transactions in the set of unlabeled data transactions are newly labeled based on the first risk score. The newly labeled data transactions are added to a second set of labeled data transactions that include the first set of labeled data transactions. A second risk determination model is trained using at least the second set of labeled data transactions. A second risk score is determined for subsequently received data transactions and these data transactions are rejected or approved based on the second risk score.

Using semi-supervised label procreation to train a risk determination model

Examples described herein generally relate to a computer device including a memory, and at least one processor configured to process a transaction. The computer device receives customer transaction information. The computer device evaluates the customer transaction information to determine a risk score for the transaction. The computer device assigns the transaction, based on the risk score, to one of a plurality of stratums including at least a first stratum and a second stratum. The computer device selects a merchant identifier (MID) from at least a first MID associated with the first stratum and a second MID associated with the second stratum based on at least the assigned stratum and a target chargeback rate for at least one of the first MID or the second MID. The computer device transmits the transaction information and the selected MID to an issuing bank.

Track bank chargeback sensitivity using disproportional risk level sampling design

Embodiments disclosed herein are related to computing systems and methods for determining a risk score for a plurality of data transactions. In the embodiments, a first risk score module may receive data transactions. The first risk score module may then determine a first risk score for each of the data transactions. A second risk score module that is different from the first risk score module may receive each of the first risk scores determined by the first risk score module as an input. The second risk score module may determine a second risk score based in part on the input first risk scores for each of the data transactions. The second risk scores may specify if each of the data transactions is to be approved or rejected by the computing system.

Long-term short-term cascade modeling for fraud detection

Reducing an amount of data transactions that are subjected to further review when determining if the data transactions should be approved or rejected. A risk score that defines a first cutoff and a second cutoff are determined. During a first time period, the first and second cutoffs are extended to increase data transactions subject to further review and to create an opportunity group of data transactions not previously approved that are now approved. A rejection rate of the data transactions subject to further review is compared to a threshold. When the rejection rate is no more than the threshold, during a second time period, a volume of the data transactions subject to further review is minimized and a second opportunity group of data transactions not previously approved that are now approved is created. The rejection rate of the data transactions subject to further review are again compared to the threshold.

Dynamic gating fraud control system

Embodiments disclosed herein are related determining a risk score for one or more data transactions. Current data transactions that are associated with one or more current attributes are received. Stored data transactions associated with stored attributes are accessed. A plurality of the stored attributes are selected. A first sliding window and a second sliding window are selected. A duration of the second sliding window is longer than a duration of the first sliding window and encompasses the duration of first sliding window. Risk information for those stored data transactions that are associated with the plurality of attributes is determined. The risk information is determined during the duration of both the first and second sliding windows and is indicative of a level fraud that is occurring. The determined risk information and the current attributes are used to generate a risk score for the current data transactions. The current data transactions are approved or rejected based on the risk score.

Hierarchical profiling inputs and self-adaptive fraud detection system

Self-learning and adaptive modeling is employed with respect to predictive analytics. A hierarchical model structure can be employed comprising a set of predictive models automatically built from accumulated data and distributed across multiple levels. For a given input type, a set of candidate models can be identified across varying levels of granularity, and a best model selected based on a comparison of performance metrics of the models. The best model can then be activated for use in making predictions. Of course, the best model can change based on most recent training performance results, since as more data becomes available more specific models can be developed.

Growing community of inventors

Sammamish, WA, United States of America

Shoou-Jiun Wang