Leveraging deep learning for automatic literature screening in intelligent bibliometrics

Intelligent bibliometrics, by providing sufficient statistical information based on large-scale literature data analytics, is promising for understanding innovative pathways, addressing meaningful insights with the assistance of expert knowledge, and indicating key areas of scientific inquiry. However, the exponential growth of global scientific publication output in most areas of modern science makes it extremely difficult and labor-intensive to analyze literature in large volumes. This study aims to accelerate intelligent bibliometrics-driven literature analysis by leveraging deep learning for automatic literature screening. The comparison of different machine learning algorithms for the automatic classification of literature regarding relevance to a given research topic reveals the outstanding performance of deep learning. This study also compares different features as model input and provides suggestions about training dataset size. By leveraging deep learning’s abilities in predictive and big data analytics, this study makes contributions to intelligent bibliometrics by promoting literature screening and is promising to track technological changes and scientific evolutionary pathways.