Journal subject classification systems are fundamental to journal evaluation, research assessments and information retrieval. Previous studies (e.g., Wang & Waltman, 2016) have identified accuracy issues in major classification systems such as the Subject Categories in the Web of Science and ASJC in Scopus. Inaccurate journal classification systems can lead to several problems such as:

➢ Possible evaluation bias, e.g., the under-representation of physical chemistry journals (Shen et al., 2024), and the suppression of LIS Journals in the JCR Information Science and Library Science category (Huang et al., 2019);

➢ Missing relevant journals in competitive analysis for journal strategic planning;

➢ Inaccurate assessments of journal disciplinary positioning leading to misguided policies (Zhang et al., 2024);

➢ Information retrieval issues (both over-inclusion and omission), affecting subsequent analyses.

Previous efforts to refine journal subject classification systems have primarily relied on journal-level citation networks (Gómez-Núñez et al., 2014; Leydesdorff et al., 2016). Now, with the emergence of accessible paper-level classification systems such as Citation Topics (Incites, 2024), we have new opportunities for improvement. For instance, Zhang and Shen (2024) demonstrated how multidisciplinary journals can be better categorized by analyzing the diversity of their covered citation topics. Furthermore, the current disconnect between subject categories and paper-level topics has constrained the full potential of paper-level classification systems. Bridging this gap would enhance classification approaches and provide a more comprehensive understanding of the scholarly literature organization.

In this initiative, we propose a hybrid approach combining quantitative and qualitative methods to refine the journal subject category and form the correspondence between subject category and paper-level topics. Figure 1 shows the data we have and the goals we want to achieve.

To achieve our goal, we will measure the journal-to-journal similarities (Zhang, 2024b) by analyzing their covered citation topics, followed by hierarchical clustering (community detection) to create a hierarchical classification system. Subject matter experts will then be involved in reviewing the results, providing interpretation, labeling the category, and fine-tuning the classification system. The complete workflow is illustrated in Figure 2. The similarity calculation forms the foundation of the classification system and the chosen similarity measurecould lead to variations in the clustering outcomes. By involving subject matter experts to review and refine the classification, we strive to minimize algorithm dependence and improve the robustness of the results.

We seek collaboration from journal editors and the research community to improve the classification system. Specifically, we invite contributors to:

➢ Report incorrectly assigned subject categories and classification errors or suggestions;

➢ Identify and recommend subject matter experts who could provide valuable insights and contribute to this project;

➢ Write and publish editorials outlining the journal landscape in specific fields and encourage broader community feedback.

Your expertise and input will be crucial in developing a more accurate and comprehensive journal subject classification system. We welcome all forms of feedback and suggestions to enhance this initiative.

The authors are from the CAS Journal Ranking teams, and the obtained journal subject classification system will be used in the CAS Journal Ranking in the near future.