He Tao, Fu Wei, Xu Jianqiao, Zhang Zhihong, Zhou Jiuxing, Yin Ying, Xie Zhenjie
Department of Information Security, Naval University of Engineering, Wuhan, China.
Front Bioeng Biotechnol. 2022 Jun 3;10:908733. doi: 10.3389/fbioe.2022.908733. eCollection 2022.
Interdisciplinary research promotes the emergence of scientific innovation. Researchers want to find interdisciplinary research in their research field. However, the number of scientific papers published today is increasing, and completing this task by hand is time-consuming and laborious. A neural network is a machine learning model that simulates the connection mode of neurons in the human brain. It is an important application of bionics in the artificial intelligence field. This paper proposes an approach to discovering interdisciplinary research automatically. The method generates an IRD-BERT neural network model for discovering interdisciplinary research based on the pre-trained model BERT. IRD-BERT is used to simulate the domain knowledge of experts, and author keywords can be projected into vector space by this model. According to the keyword distribution in the vector space, keywords with semantic anomalies can be identified. Papers that use these author keywords are likely to be interdisciplinary research. This method is applied to discover interdisciplinary research in the deep learning research field, and its performance is better than that of similar methods.
跨学科研究促进了科学创新的出现。研究人员希望在其研究领域中找到跨学科研究。然而，如今发表的科学论文数量不断增加，手动完成这项任务既耗时又费力。神经网络是一种模拟人脑神经元连接方式的机器学习模型。它是仿生学在人工智能领域的一项重要应用。本文提出了一种自动发现跨学科研究的方法。该方法基于预训练模型BERT生成了一个用于发现跨学科研究的IRD-BERT神经网络模型。IRD-BERT用于模拟专家的领域知识，通过该模型可以将作者关键词投影到向量空间中。根据向量空间中的关键词分布，可以识别出语义异常的关键词。使用这些作者关键词的论文很可能是跨学科研究。该方法被应用于深度学习研究领域中发现跨学科研究，其性能优于类似方法。
