Chen Yifei, Li Xiaoying, Li Aihua, Li Yongjie, Yang Xuemei, Lin Ziluo, Yu Shirui, Tang Xiaoli
Institute of Medical Information, Chinese Academy of Medical Sciences, Beijing, China.
JMIR Form Res. 2023 Aug 18;7:e47434. doi: 10.2196/47434.
The normalization of institution names is of great importance for literature retrieval, statistics of academic achievements, and evaluation of the competitiveness of research institutions. Differences in authors' writing habits and spelling mistakes lead to various names of institutions, which affects the analysis of publication data. With the development of deep learning models and the increasing maturity of natural language processing methods, training a deep learning-based institution name normalization model can increase the accuracy of institution name normalization at the semantic level.
This study aimed to train a deep learning-based model for institution name normalization based on the feature fusion of affiliation data from multisource literature, which would realize the normalization of institution name variants with the help of authority files and achieve a high specification accuracy after several rounds of training and optimization.
In this study, an institution name normalization-oriented model was trained based on bidirectional encoder representations from transformers (BERT) and other deep learning models, including the institution classification model, institutional hierarchical relation extraction model, and institution matching and merging model. The model was then trained to automatically learn institutional features by pretraining and fine-tuning, and institution names were extracted from the affiliation data of 3 databases to complete the normalization process: Dimensions, Web of Science, and Scopus.
It was found that the trained model could achieve at least 3 functions. First, the model could identify the institution name that is consistent with the authority files and associate the name with the files through the unique institution ID. Second, it could identify the nonstandard institution name variants, such as singular forms, plural changes, and abbreviations, and update the authority files. Third, it could identify the unregistered institutions and add them to the authority files, so that when the institution appeared again, the model could identify and regard it as a registered institution. Moreover, the test results showed that the accuracy of the normalization model reached 93.79%, indicating the promising performance of the model for the normalization of institution names.
The deep learning-based institution name normalization model trained in this study exhibited high accuracy. Therefore, it could be widely applied in the evaluation of the competitiveness of research institutions, analysis of research fields of institutions, and construction of interinstitutional cooperation networks, among others, showing high application value.
机构名称的规范化对于文献检索、学术成果统计以及研究机构竞争力评估至关重要。作者写作习惯的差异和拼写错误导致机构名称多种多样，这影响了出版数据的分析。随着深度学习模型的发展和自然语言处理方法的日益成熟，训练基于深度学习的机构名称规范化模型可以提高语义层面机构名称规范化的准确性。
本研究旨在基于多源文献的机构隶属数据特征融合，训练一种基于深度学习的机构名称规范化模型，借助权威文件实现机构名称变体的规范化，并经过多轮训练和优化达到较高的规范准确率。
在本研究中，基于来自变换器的双向编码器表征（BERT）和其他深度学习模型，训练了一个面向机构名称规范化的模型，包括机构分类模型、机构层次关系提取模型和机构匹配与合并模型。然后通过预训练和微调对该模型进行训练，以自动学习机构特征，并从3个数据库（Dimensions、科学引文索引和Scopus）的机构隶属数据中提取机构名称，完成规范化过程。
发现训练后的模型至少可以实现3个功能。首先，该模型可以识别与权威文件一致的机构名称，并通过唯一的机构ID将该名称与文件关联起来。其次，它可以识别非标准的机构名称变体，如单数形式、复数变化和缩写，并更新权威文件。第三，它可以识别未注册的机构并将其添加到权威文件中，这样当该机构再次出现时，模型可以识别并将其视为注册机构。此外，测试结果表明，规范化模型的准确率达到93.79%，表明该模型在机构名称规范化方面具有良好的性能。
本研究训练的基于深度学习的机构名称规范化模型具有较高的准确率。因此，它可广泛应用于研究机构竞争力评估、机构研究领域分析以及机构间合作网络构建等方面，具有较高的应用价值。
