Machine learning predictive modeling with natural language processing of end-user communications

Abstract

Disclosed is an approach for training and applying predictive models that employ natural language processing in analyzing records with summary and resolution fields. A semi-supervised machine learning text model may be used to generate a category score for each record in a first dataset. The model may be generated using a training dataset comprising past records and categorization labels for each of the past records. The model may perform textual analytics on summary and resolution fields received as inputs and provide category scores as output. Dispositions may be assigned to each record in the first dataset based on a comparison of the category score to a category threshold. A second dataset may be generated with the records of the first dataset and assigned dispositions. A report comprising records and, for each record, disposition and category score may be generated.