Combining natural language processing and multidimensional classifiers to predict and correct CMMS metadata

Computerized maintenance management systems (CMMSs) contain valuable data on the operations in an organization. A large part of these consists unstructured, written texts contained failure notifications which are generated each time unexpected occurs, enriched with structured metadata consisting a number labels that allow to categorize failures, such as type failure, its cause or corrective action was taken. In this paper, we show natural language processing techniques can be used predict based unstructured text and even identify mislabeled ambiguous labels. Specific attention is given complexity arises from highly technical nature combined telegraphic writing style heavy use sentence fragments abbreviations. Moreover, it shown exploiting dependencies between different components metadata, regarding prediction problem multidimensional classification problem, improve reliability predicted We illustrate test our label pipeline CMMS pharmaceutical company.