Discovering Interdisciplinary Research Based on Neural Networks.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computational Linguistics

Background:

• Scientific innovation often arises from interdisciplinary research.
• Identifying interdisciplinary research manually is challenging due to the vast volume of publications.
• Automated methods are needed to efficiently discover cross-field scientific work.

Purpose of the Study:

• To develop an automated approach for discovering interdisciplinary research.
• To propose a novel neural network model for identifying papers that bridge different scientific domains.
• To enhance the efficiency of researchers in finding relevant interdisciplinary studies.

Main Methods:

• A neural network model, IRD-BERT, was developed based on the BERT pre-trained model.
• IRD-BERT projects author keywords into a vector space to simulate expert domain knowledge.
• Semantic anomalies in keyword distribution are identified to flag potential interdisciplinary papers.

Main Results:

• The IRD-BERT model successfully identified interdisciplinary research within the deep learning field.
• The proposed method demonstrated superior performance compared to existing approaches for detecting interdisciplinary papers.
• Analysis of keyword vector space revealed semantically anomalous terms indicative of cross-disciplinary work.

Conclusions:

• The IRD-BERT model offers an effective and automated solution for discovering interdisciplinary research.
• This approach can significantly reduce the manual effort required by researchers to find innovative, cross-field studies.
• The method holds promise for advancing scientific discovery by facilitating the identification of emerging research trends.