Domain Lexicon Enhanced Named Entity Recognition for Industrial Maintenance Field

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

With the increasing complexity of industrial maintenance scenarios, efficiently and accurately extracting key information from the text of maintenance work orders has become a critical challenge that needs to be solved in the field of maintenance operations. In order to help maintenance personnel identify the key information in the work order, this paper proposes a named entity recognition(NER) method based on a domain-lexicon-enhanced Bidirectional Encoder Representations from Transformers - Conditional Random Field (BERT-CRF). In the data processing stage, the dataset was divided into training, validation, and test sets using a fixed random seed of 42. For unstructured text data, this paper builds a NER model that integrates professional lexicon information in the industrial field to optimize the extraction effect of key entities. Complete the entry matching through the longest matching, concatenate and fuse the lexicon feature embedding with the Bidirectional Encoder Representations from Transformers (BERT) context vector and combine with the Conditional Random Field (CRF) layer learning label sequence constraints to realize the entity recognition of industrial maintenance work orders. The experimental results show that the Micro-F1 of the lexicon-enhanced BERT-CRF model reaches 93.57%, which is 1.98 percentage points higher than the BERT-CRF baseline, indicating that the domain lexicon can effectively improve the overall entity recognition performance on industrial maintenance work order texts.