A Deep-Learning Approach to Determine the Dependency between Two Subject Types in the Web of Science

Abstract

Traditional wisdom always suggests that some subjects have strong relationships, while others are almost mutual independent. However. there is a lack of quantitative approach to formulate these relationships in a systematic and unbiased way. In this work, we train a classification machine via deep learning to determine whether two subject types are independent based on the citation information from the Web of Science database. This machine not only achieves very high accuracy in estimating the dependency among subject types in the database, but also is able to predict the dependency when one or both subject types do not exist in the database.