How NLP and Visual Analytics Can Improve Asset Management

In Asset Management, maintenance work is mainly reported through Work Orders (WO), which are technical documents that specify the asset to be repaired. These works are described in free text, with no imposed structure or fixed vocabulary, which makes them difficult to analyse automatically. This challenge becomes more significant as the number and variety of assets increase. This study presents the use of Natural Language Processing (NLP) to automate the processing of work order descriptions. NLP algorithms can summarise large amounts of text into concise summaries. To understand the text corpus and effectively communicate the results to understand the underlying semantic patterns better, two well-known Word Embeddings (WE) models, Word2Vec and Fasttext, capture the semantic and syntactic relationships between words. By reducing the dimensions of the encoded vectors, it becomes possible to explore a 3D vector space through vector visualisation interactively. The results show that the Fasttext approach outperforms Word2Vec in capturing semantic information. This allows the development of machine learning algorithms to summarise a work order using a small set of predefined words.