Unsupervised Production Machinery Data Labeling Method Based on Natural Language Processing
Makar Savchenko, В С Тынченко
Abstract
This paper proposes an approach for the unsupervised labeling of production machinery data with the leverage of Natural Language Processing (NLP) techniques. Acquiring labeled data for training machine learning models poses a significant challenge involving high monetary and time costs, which can be tremendously reduced using the method presented in this paper. The method, described in this paper, uses standard NLP techniques in combination with vector representations of dataset object descriptions, obtained using a pre-trained transformed-based model, to automate the dataset markup. The proposed method was evaluated on a real-world production machinery dataset and demonstrated its effectiveness in assigning meaningful labels in a natural language to dataset records. Implications of this work can be extended to any sector, having classes and descriptions of dataset objects in a natural language.