Abstract

As interdisciplinary research expands through the convergence of academic fields and the number of accessible electronic journals increases, researchers face growing challenges in selecting appropriate journals for manuscript submission. There is a lack of research on journal recommendation systems that reflect the Korean academic ecosystem, in which academic services offer different sets of journals and international journals published by Korean academic societies are increasing. This study proposes a machine learning-based journal recommendation architecture that leverages language models. The proposed architecture embeds paper titles and abstracts using BERT-based language models further trained on target data, and these embedded vectors are then input into an XGBoost classifier to recommend appropriate journals. Analysis results showed that among BERT-based models, RoBERTa demonstrated the best performance, with its recommendation system outperforming approximately 13% higher compared to systems based on traditional natural language processing techniques. Furthermore, it was found that recommendations for papers outside the scope of service journals and papers written in Korean were feasible. This study contributes both academically and practically by presenting an academic journal recommendation architecture that leverages language models and machine learning by considering the actual Korean academic publishing environment.