基于LSTM的桥梁养护文本数据的命名实体识别方法

LSTM-based Named Entity Recognition Method for Bridge Maintenance Text Data

为高效抽取桥梁养护文本数据的关键实体信息,提升养护工作的效率和质量,提出基于LSTM的桥梁养护命名实体识别方法。为实现上述方法,在数据标注、人工智能模型可用性和模型架构优化方面进行研究和验证。在数据标注方面提出结合桥梁养护四要素标注原则,以有效提取桥梁养护文本中的核心信息。在人工智能模型可用性方面,在桥梁养护标注文本数据上测试基于LSTM的一系列命名实体识别模型,以验证人工智能模型在桥梁养护领域内的可用性。在模型架构优化方面,模型通过Dense Internal和Attention模块及超参数调整取得更优性能。这三方面的研究和验证工作为后续研究提供了数据标注的参考原则,人工智能模型引入桥梁养护领域的可用性依据和桥梁养护领域内架构优化的潜在方向。

To efficiently extract the key entity information from bridge maintenance text data,and improve the efficiency and quality of maintenance work,the LSTM-based named entity recognition method for bridge maintenance was proposed.To achieve the above method,the study and verification were conducted in data annotation,artificial intelligence model availability,and model architecture optimization.In terms of data annotation,the annotation principle of combining 4 elements of bridge maintenance was proposed to effectively extract core information from bridge maintenance text.For the artificial intelligence model availability,a series of LSTM-based named entity recognition models was tested on annotated bridge maintenance text data to verify the availability of artificial intelligence models in bridge maintenance field.For the model architecture optimization,the model achieved better performance by using Dense Internal and Attention modules and hyperparameters tuning.The study and verification work in the above 3 aspects provide the reference principles of data annotation for subsequent research,the availability basis for introducing artificial intelligence models into bridge maintenance field,and the potential directions for architecture optimization in bridge maintenance field.