基于语义块识别的材料科学文献工艺数据实体关系抽取  

Semantic Block Labeled for the Process Data Entity Relation Extraction in Materials Science Literature

作　　者：黄星瑞 HUANG Xing-rui(Faculty of Information Engineering and Automation,Kunming University of Science and Technology)

机构地区：[1]昆明理工大学信息工程与自动化学院

出　　处：《化工自动化及仪表》2024年第3期507-515,共9页Control and Instruments in Chemical Industry

摘　　要：提出一种语义块标注的材料科学文献工艺实体关系抽取方案和基于多步实体识别的流水线式材料科学文献工艺实体关系抽取模型。针对材料科学文献中工艺数据的语义逻辑块通常连续出现的特点,首先采用语义块标注方案将工艺文本作为一个序列完整地进行命名实体识别,然后将提取出的工艺数据序列单独进行分类和进一步命名实体识别,最后依据类型分类结果和实体识别结果对工艺操作与条件参数构建关系三元组。实验结果表明,相较于基线算法,所提标注方案在工艺实体识别上的准确度提升了4%,所提模型在工艺数据实体关系抽取实验中F1分数提升了3.6%。A semantic block labeled method for the process entity relationship extraction in materials science literature was proposed,including the pipelined entity relation extraction model for materials science literature based on multi-step entity recognition.Aiming at the characteristic that the semantic logic blocks of process data usually appear continuously in materials science literature;firstly,having a semantic block annotation scheme used to take process text as a sequence for named entity recognition;and then,having the extracted process data sequence classified separately and the entity recognition further named;and finally,having the relationship triples between process operations and condition parameters constructed according to both type classification results and entity recognition results.Experimental results show that,compared to the baseline algorithm,the accuracy of the proposed annotation scheme in process entity recognition is improved by 4%,and the Fl score of the proposed model in the process data entity relationship extraction experiment is increased by 3.6%.

关 键 词：命名实体识别 关系抽取 材料工艺数据 科学文献 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]