A deep learning approach to enhance accuracy and diversity of recommendation for interdisciplinary journals

Abstract

To meet scholars' need to recommend both higher accuracy and diversity when submitting interdisciplinary papers, this paper proposes an improved journal diversity recommendation method based on the attention mechanism in deep learning. This method can retain all key information in long texts by using the attention mechanism. It identifies and stores the research directions and hotspots covered in different papers across journals to extract common research topics for each journal type. Five deep learning models based on attention mechanism are introduced, 104,176 paper abstracts from 111 Web of Science journals are used to fine-tune the models. After learning on training set and model testing on the test set, recommendation accuracy and diversity results are calculated for 9 categories. Finally, the recommendation accuracy and diversity of the 5 attention mechanism based deep learning models are compared with benchmark models across different journal types. The experimental results demonstrate the feasibility and superiority of this method comprehensively considering the metrics of accuracy and diversity at a large scale. It provides theoretical and practical advancements to develop an effective journal recommender system which helps scholars to make wise decision for journal submission.