Fusion Features-Based Entity Recognition Method for Safety Knowledge of Non-Coal Open-Pit Mine

Knowledge graph technology that provides important information and data support for improving the level of safety production, brings together related laws, regulations and construction methods of non-coal open-pit mining production. However, as a key step in the construction of knowledge graph, it is a major challenge to recognize and extract entities from the complex field of safety production in non-coal open-pit mines. In this paper, a new entity recognition method based on fusion features, MSAL (Multilayer Self-attention Lexicon), is proposed, which shows better performance of entity recognition in this special field. A word-level enhancement feature SoftLexicon is adopted to solve the problem of flat entity boundary generated by character sequence model in Chinese named entity recognition. Then, the SoftLexicon feature information is dynamically weighted and fused using a self-attention mechanism. In order to solve the problem of text information not being fully utilized by the pre-trained model, a multi-layer fusion method combining hidden state and transformer layers is proposed. Comparison and ablation experiments were carried out to demonstrate the proposed method’s effects. The experimental results show that the recall rate of relevant indicators under the test set is 67.54% in contrast to other models, and the training speed and portability performance are obviously better.