Abstract: Certain aspects of the present disclosure provide techniques for performing inferences in a distributed computing environment. An example method generally includes receiving a request to perform an inference with respect to a set of input data. One or more client devices are selected for use in performing the inference with respect to the set of input data. A request to perform the inference is transmitted to the selected one or more client devices. The request generally includes an anonymized, vectorized version of the set of input data such that the selected one or more client devices are to perform the inference based on anonymized data. An inference response is received from each of the selected one or more client devices.

Abstract: Systems and methods that may be used to automatically generate inventory templates for use with an accounting platform. The automatically generated templates may be for a first user within a particular industry and may be based on established inventory trees of other system users within the same industry that have similar demographics of the first user.

Abstract: A method of zoning a transcription of audio data includes separating the transcription of audio data into a plurality of utterances. A that each word in an utterances is a meaning unit boundary is calculated. The utterance is split into two new utterances at a work with a maximum calculated probability. At least one of the two new utterances that is shorter than a maximum utterance threshold is identified as a meaning unit.

Abstract: Systems and methods that may be used to generate and use a social graph generated by user financial transaction data (i.e., a financial transaction-based social graph). Connections and other data within the financial transaction-based social graph can be used for targeted product offerings, other offerings, and or advertisements via e.g., collaborative filtering and user segmentation and profiling.

Abstract: A method of training machine learning models (MLMs). An issue vector is generated using an issue MLM to generate a first output including first embedded natural language issue statements. An action vector is generated using an action MLM to generate a second output including related embedded natural language action statements. The issue and action MLMs are of a same type. An inner product of the first and second output is calculated, forming a third output. The third output is processed according to a sigmoid gate process to predict whether a given issue statement and corresponding action statement relate to a same call, resulting in a fourth output. A loss function is calculated from the fourth output by comparing the fourth output to a known result. The issue MLM and the action MLM are modified using the loss function to obtain a trained issue MLM and a trained action MLM.

Abstract: Automatic keyphrase labeling and machine learning training may include a processor extracting a plurality of keywords from at least one search query that resulted in a selection of a document appearing in a search result. For each of the plurality of keywords, the processor may determine a probability that the keyword describes the document. The processor may generate one or more keyphrases by performing processing including selecting each of the plurality of keywords having a probability greater than a predetermined threshold value for insertion into at least one of the one or more keyphrases and assembling the one or more keyphrases from the selected plurality of keywords. The processor may label the document with the keyphrase.

Abstract: A method collects word-based data corresponding to a first identifier. A first phrase vector is generated for the first identifier by extracting frequency data from the word-based data. A similarity metric is generated corresponding to the first identifier and a second identifier by comparing the first phrase vector of the first identifier to a second phrase vector of the second identifier. A tuple is generated that includes the first identifier and the second identifier using the similarity metric. A machine learning model is trained with the tuple to generate an embedded vector corresponding to the first identifier.

Abstract: A method may include detecting, in transactions of initial users, open recurring expense sequences each having expense sequence attributes, deriving, using the expense sequence attributes of the open recurring expense sequences, recurring expense groups each including a subset of the initial users, generating a prediction that the open recurring expense sequences of a recurring expense group will terminate within a period of a current period, grouping, using personal attributes of the users in the recurring expense group, the recurring expense group into recurring expense subgroups, generating, using a trained model, scores for the recurring expense subgroups each indicating a probability that the open recurring expense sequences of the respective recurring expense subgroup are extendable beyond the current period, and selecting, using the scores for the recurring expense subgroups, a recurring expense subgroup to attempt an extension of the open recurring expense sequences of the recurring expense subgroup.

Abstract: Certain embodiments of the present disclosure provide techniques for detecting anomalous activity in a computing system. The method generally includes receiving a request to perform an action in a computing system. The request is added to a historical time-series data set. A portion of the historical time-series data set is selected for use in determining whether the received request is an anomalous request, and a set of previously identified outliers are removed from the selected portion of the historical time-series data set. An anomaly score is calculated based on a statistical analysis of the received request and the selected portion of the historical time-series data set, wherein the anomaly score comprises a predicted number of operations executed to isolate the received request from the selected portion of the historical time-series data set. One or more actions are taken to process the received request based on the calculated anomaly score.

Abstract: This disclosure relates to systems and methods for constructing a customized debt reduction plan for a user. In some implementations, a customized debt reduction system obtains a plurality of financial attributes of the user and a plurality of other users, where the plurality of financial attributes are indicative of credit card debt, and identifies users from the plurality of other users who successfully repaid their credit card debt based on their respective financial attributes and one or more repayment techniques that resulted in successful repayment of their credit card debt. The customized debt reduction system correlates the plurality of financial attributes of the user with the plurality of financial attributes of a number of the identified users and determines a personalized score for the user, using a trained machine learning model, based on the correlation to determine a customized debt reduction plan for the user based on the personalized score.

Abstract: A method of zoning a transcription of audio data includes separating the transcription of audio data into a plurality of utterances. A that each word in an utterances is a meaning unit boundary is calculated. The utterance is split into two new utterances at a work with a maximum calculated probability. At least one of the two new utterances that is shorter than a maximum utterance threshold is identified as a meaning unit.

Abstract: Improved systems and method as disclosed herein, provide automated analysis tools for more refined trend analysis and evaluation of identified trends. Communication data may be recognized as either audio or textual data which may be processed and analyzed in real-time (as in the case of streaming audio data) or processed at a time apart from the acquisition of the communication data. If the communication data is audio data, then the audio data, may undergo a transcription, which may employ the exemplary technique of large vocabulary continuous speech recognition (LVCSR) or other known speech-to-text algorithms or techniques. Alternatively, the communication data may already be in the form of a transcription or the communication data may have originated as textual data, exemplarily the communication data is from an internet web chat, email, text message, or social media.

Abstract: Systems and methods that may be used to provide personalized financial nudges to users of a financial service that may be used to further the users' savings intentions (e.g., a savings goal, an emergency fund, etc.). The disclosed systems and methods may increase user interactivity with the financial service and the services it offers by providing personalized nudges that are based on, among other things, an evaluation of various behavioral economics principles. A machine learning recommendation system may be used to fit and output different nudges to users in a personalized way to maximize their savings' intentions.

Abstract: Certain aspects of the present disclosure provide techniques for predicting a location based on transaction record data. An example technique includes obtaining a first set of transaction records and determining a merchant associated with each transaction record. The example further includes based on the merchant, determining and appending a branch identifier to each transaction record associated with the merchant to generate a first set of extended transaction records. The example further includes creating a consumption graph based on the first set of extended transaction records and determining an estimated location based on the consumption graph. The example further includes determining a precise point location based on the estimated location.

Abstract: A method for improving a transcription may include identifying, in the transcription, reliable channel tokens of an utterance of a reliable channel and an unreliable channel token of an utterance of an unreliable channel, and generating, using a machine learning model, a vector embedding for the unreliable channel token and vector embeddings for the reliable channel tokens. The method may further include calculating vector distances between the vector embedding and the vector embeddings, and generating, for the unreliable channel token and using the vector distances, a score corresponding to a reliable channel token. The method may further include determining that the score is within a threshold score, and in response to determining that the score is within the threshold score, replacing, in the transcription, the unreliable channel token with the reliable channel token.

Abstract: Aspects of the present disclosure provide techniques for database documentation propagation. Embodiments include scanning a log comprising a plurality of database queries to identify one or more database queries of the plurality of database queries, the one or more database queries being associated with generating a new table of a database based on information in an existing table of the database. Embodiments include generating, based on the one or more database queries identified during the scanning, a directed acyclic graph (DAG) comprising: a first vertex representing the existing table; a second vertex representing the new table; and a directed edge connecting the first vertex to the second vertex. Embodiments include obtaining documentation associated with the existing table. Embodiments include propagating, based on the DAG, at least a subset of the documentation associated with the existing table to the new table.

Abstract: A method and system identify characteristics of financial transaction description strings. The method and system trains an analysis model with a machine learning process to classify financial transaction description strings. The analysis model generates a table that indicates the portions of the financial transaction description strings that were relevant in classifying the financial transaction description strings and the portions that were not relevant.

Abstract: A method may include receiving, for a package, shipment details including attributes, obtaining, for a subset of the attributes, logistic preferences, applying the logistic preferences to the shipment details to obtain modified shipment details, training a classifier using shipment transactions each including values for the attributes and labeled with a vendor logistic service, generating, by applying the classifier to the modified shipment details, scores for vendor logistic services, and recommending a vendor logistic service from the vendor logistic services using the scores.

Abstract: A method may include generating, using a flow proportionalized graph, scores for platform sellers of an online platform. The flow proportionalized graph may include nodes corresponding to the platform sellers and buyers, and edges each connecting a buyer node corresponding to a buyer initiating a monetary transfer and a platform seller node corresponding to a platform seller receiving the monetary transfer. Each edge may have a weight that is a proportion of total monetary transfers by the buyer received by the platform seller. The method may further include matching, using the scores and a seller similarity metric, a non-platform seller with a platform seller, receiving a scenario for the platform seller to sell an item of the non-platform seller via the online platform, and generating a prediction regarding an outcome of the scenario by applying a model to scenarios.

Abstract: Systems and methods that may be used to provide a predictive tax loan or other monetary advance before the loan recipient (e.g., a taxpayer) prepares and files its tax return. A risk of providing a predictive tax loan or monetary advance is modeled separately from a machine learning model used to determine the anticipated tax refund amount and tax loan. The disclosed systems and methods may also predict accurate tax withholdings based on multiple machine learning models from multiple services, including non-payroll related services.