融合词汇边界信息的合同实体识别方法
Contract entity recognition method with lexical boundary information
王浩畅 1和婷婷 1郑冠彧1
作者信息
- 1. 东北石油大学计算机与信息技术学院,黑龙江大庆 163318
- 折叠
摘要
针对合同中实体表达形式复杂多变、识别粒度细的特点,及合同文本中实体较长问题,提出一种融合词汇边界信息的合同实体识别方法.利用预训练语言模型动态生成语义向量作为模型输入;运用相对位置编码对Transformer结构进行改进,使其在编码过程中融合词汇信息,进一步丰富语义特征;通过条件随机场(CRF)结构进行解码,得到输入序列的标签预测.实验结果表明,该方法可以有效确定合同文本中的实体边界,具有良好的泛化性能.
Abstract
To solve the problems of the complex and variable entity expression form in contract,the fine recognition granularity,and the long entity in contract texts,a contract entity recognition method based on lexical boundary information was proposed.The pre-trained language model was used to dynamically generate semantic vectors as model input.The relative position encoding was used to improve the Transformer structure to integrate lexical information in the encoding process and further enrich the semantic features.The conditional random field(CRF)structure was used for decoding,and the label prediction of the input sequence was obtained.Experimental results show that the proposed method can effectively recognize the entity boundary in the contract texts and has good generalization performance.
关键词
实体识别/合同文本/预训练语言模型/相对位置编码/转换器结构/词汇边界信息/条件随机场Key words
entity recognition/contract text/pre-trained language models/relative position encoding/transformer structure/lexical boundary information/conditional random field引用本文复制引用
基金项目
国家自然科学基金(61402099)
国家自然科学基金(61702093)
出版年
2024
NETL
NSTL
万方数据