Recommendation System for Journals based on ELMo and Deep Learning

Abstract

Choosing the right journal to publish research studies is critical for researchers. Despite its importance, the task of determining the suitable and high-ranking journal for publishing can be challenging due to several factors such as the growing number of the available journals and the fact that every journal has its specific area of expertise. In this paper we investigate content-based journal recommendation systems that rely on using NLP to analyze features of existing journals and use those to pre-select a particular number of suitable journals for a new paper. Our experiments are based on the ELMo feature engineering mechanism and use different deep learning neural network architectures (CNN, RNN). We used datasets from the disciplines physics, chemistry and biology, with each containing data of more than 750000 publications. The data source consists of the abstracts of the papers. The experiments show promising results with the accuracy of our models outperforming existing models. Specffically, our RNN model can achieve 83% accuracy when using data from physics by considering top-20 journals.