Leveraging Semantic Similarity for Non-Conformity management in industrial environments

In this work we explore the use of semantic similarity techniques to enhance the management of Non-Conformity in emerging industrial environments. Small and new companies lack historical data which makes unfeasible the application of Machine Learning-based solutions for data processing. Regarding this, we investigate the usage of training-independent semantic similarity techniques for Non-Conformity matching. Specifically, we compared two text vectorization methods, Term Frequency-Inverse Document Frequency and the Word2vec model combined with the cosine distance algorithm to reduce the amount of Non-Conformity to be processed by human specialists automatically finding the reports that point out the same quality issue.