Abstract: The concept involves efficiently using machine learning to quickly identify possible fraudulent applications in small business loan and credit applications by automatically flagging applications that meet certain criteria. In one preferred implementation, the tool compares a business description to a selected NAICS code in a loan application to assess the potential for fraud. Specifically, an algorithm can match the leftmost two digits of the selected code with the description of the category from an applicant. An engine calculates a probability of a fraud score based on the matching attached to the application. Because the tool detects fraud proactively rather than reactively, it substantially reduces computational costs and resources and reduces the biases associated with highly intensive manual work.

Abstract: Various examples are directed to systems and methods for visualizing computerized model decisions. A model system may access result data that comprises event results for a plurality of events and event feature data describing sets of event features. A first event result for a first event of the plurality of events is generated using the computerized model and a first set of event features from the event feature data. The model system may access importance score data comprising importance scores for at least a portion of the first set of event features, wherein a first event feature is assigned a first importance score and a second event feature is assigned a second importance score indicating a greater importance than the first importance score. The model system may also generate a user interface for display to a human user. The user interface may comprise a background area; a first visual indicator positioned on the background area and corresponding to the first event feature.

Abstract: Techniques are described for evaluating message content in each of a plurality of messages in order to improve communication between customers and customer service representatives. The disclosed techniques enable efficient evaluation of the message content in each of the messages in order to determine a readability score for each of the messages. A computing device performs the efficient evaluation by processing the message content into a binary representation using a first value to represent vowels and a second value to represent non-vowels. The binarization of the message content allows the computing device to perform a syllable count and determine a readability score for each of the plurality of messages, which may include millions of messages. Based on the readability scores, the computing device may identify readability gaps between the representative messages and the customer messages, and generate subsequent representative messages that reduce the identified readability gaps.

Abstract: Systems and methods create and modify augmented deal structure using specialized components working together in a technical system by aggregating and transforming inputs. Big Data issues are controlled and restriction rules along with specialized components work together to synthesize an augmented deal structure. This augmented structure is presented through a graphical user interface configured to provide at least one of an interactive analysis (providing sensitivity analytics) and a batch processing capability.

Abstract: This disclosure describes techniques that include predicting a business metric associated with a given business enterprise using machine learning algorithms that have been trained on data, from other businesses, about that same business metric. In one example, this disclosure describes a method that includes selecting, from a data set, a training data subset and a test data subset; generating a plurality of candidate models configured to generate a predicted output metric, wherein the plurality of candidate models is generated by applying a plurality of machine learning algorithms to the training data subset; evaluating, using the test data subset, each of the plurality of candidate models to select a chosen model; receiving production data that includes data representing input metrics for a business entity not included within the plurality of business entities; and generating a predicted output metric for the business entity.

Abstract: Various examples are directed to systems and methods for visualizing computerized model decisions. A model system may access result data that comprises event results for a plurality of events and event feature data describing sets of event features. A first event result for a first event of the plurality of events is generated using the computerized model and a first set of event features from the event feature data. The model system may access importance score data comprising importance scores for at least a portion of the first set of event features, wherein a first event feature is assigned a first importance score and a second event feature is assigned a second importance score indicating a greater importance than the first importance score. The model system may also generate a user interface for display to a human user. The user interface may comprise a background area; a first visual indicator positioned on the background area and corresponding to the first event feature.

Abstract: Techniques are described for evaluating message content in each of a plurality of messages in order to improve communication between customers and customer service representatives. The disclosed techniques enable efficient evaluation of the message content in each of the messages in order to determine a readability score for each of the messages. A computing device performs the efficient evaluation by processing the message content into a binary representation using a first value to represent vowels and a second value to represent non-vowels. The binarization of the message content allows the computing device to perform a syllable count and determine a readability score for each of the plurality of messages, which may include millions of messages. Based on the readability scores, the computing device may identify readability gaps between the representative messages and the customer messages, and generate subsequent representative messages that reduce the identified readability gaps.