Named Entity Recognition in Industrial Processes

Natural Language Processing (NLP) tasks like relation extraction and knowledge graph are based on named entity recognition (NER). In order to improve the recognition ability of Chinese entities in industrial processes, a NER model based on BiLSTM-CRF network was proposed. Firstly, the abstract of patent of industrial processes was crawled from the Internet through crawler. After data cleaning, de-duplication and coding, it became a data set. Then the data was put into BiLSTM for bidirectional coding to obtain long sequence semantic features, which can be decoded through Conditional Random Field (CRF). By learning the dependency between tags, it obtain the optimal tag sequence. Finally, the correct entity was identified by sequence. In the self built dataset of industrial processes, the precision, recall and F1-score of the model are 97.17%, 99.41% and 98.27% respectively, which can prove the model can effectively improve the Chinese entity recognition ability of industrial processes.