Learning to recommend journals for submission based on embedding models

Chao Liu,Xizhao Wang,Han Liu,Xiaoying Zou,Si Cen,Guoquan Dai

DOI: https://doi.org/10.1016/j.neucom.2022.08.043

IF: 6

2022-01-01

Neurocomputing

Abstract:Due to the rapid development of electronic journals, selecting appropriate journals to publish research papers has become a significant challenge to researchers. Sometimes, even a high-quality paper may get rejected from the editor due to the mismatch between the topic of the paper and the scope of the journal. To address this issue, we present a framework of learning to recommend journals for submission based on embedding models to assist researchers in journal selection. Specifically, the journal recommen-dation problem is formulated in the context of multi-class classification, where the Bidirectional Encoder Representations from Transformers (BERT) is deployed to extract the text-level features of representing papers and the AutoEncoder (AE) network is adopted to obtain the feature representation of each journal from the relationship matrix of the paper-journal bipartite graph. The final recommendation of journals is made by using a scoring function and a Softmax classifier. Experimental results obtained on the closed dataset of 10 different journals and the DBLP dataset indicate that we proposed method outperforms sev-eral classical approaches in terms of accuracy, F1, MRR, etc. Furthermore, we introduce information entropy as an evaluation index and analyze the model performance from the perspective of prediction uncertainty. This study provides a new approach to the journal recommendation task, and researchers can choose the appropriate embedding methods according to the actual problem.(c) 2022 Elsevier B.V. All rights reserved.