Research on named entity recognition method in ship diesel engine fault field

Knowledge graph technology has the ability to integrate unstructured data and utilize prior knowledge of ship diesel engine faults. Therefore, a named entity recognition method based on BiLSTM CRF fusion of entity boundary information (EBA-NER) is proposed in the paper. Firstly, a fine-tuning BERT model is used to encode the words in the prior knowledge sequence of ship diesel engine faults. Then, the BiLSTM model is used to extract the feature relationships between words. Then, the entity boundary perception module and CRF layer are used to label the words of each entity boundary in the training set with various types of special labels. By combining entity boundary correlation analysis, the potential association between entity boundary information and entity categories is analyzed to improve the accuracy of the model's judgment of entity boundaries and categories. The CRF model corrects the output of the BiLSTM layer by learning the transition probability between labels in the dataset. The experimental results show that the method outperforms other methods such as BiLSTM CRF in named entity recognition performance in prior knowledge of ship diesel engine faults.