Our goal is to extract entities from the text data of unstructured CNC machine tool design for the construction of knowledge graph. The key entity extraction problem in the construction of CNC machine tool design knowledge graph is studied. In order to realize the recognition of named entities, we have formulated the standard and labeling method of knowledge classification for the field of CNC machine tools, and constructed the corresponding domain data set. In addition, we also propose an entity recognition technology based on RoBertTa-BiLSTM-LCRF for CNC machine tool design text. Firstly, we fine-tune the RoBertTa-BiLSTM-LCRF model using data sets in the field of CNC machine tools, and then use RoBERTa to encode the text to generate a vector representation; next, we use bidirectional long short-term memory (BiLSTM) to extract the features of vectors. Finally, we introduce LCRF as the overall optimization layer of the label, so as to derive the best answer and label the entity.The experimental results show that the F1 value of the model in the data set reaches 71.16 %; for most of the key entities, the value of F1 exceeds 65 %; this method shows significant advantages in the entity recognition of CNC machine tool design knowledge. It can accurately identify the core entities in the machine tool design knowledge document, and provides a solid data support for the construction of CNC machine tool design knowledge graph.
