A term function-aware keyword citation network method for science mapping analysis

Various keyword network methods are used to map scientific fields, but few studies have considered the semantic roles of keywords in such networks. This study proposes a term func-tion-aware keyword citation network to fill this research limitation. Specifically, we first used a term function identification method to identify research questions and methods from scientific articles. Then, we constructed a question-method term citation network to represent the corre-lation structure of keywords. Next, we explored the topology characteristics, question-method bipartite network, and knowledge community structure of the generated network to validate its superiority in science mapping analysis. A dataset of 299,567 conference proceedings collected from the Association for Computing Machinery (ACM) digital library is used to evaluate the effectiveness of our methods. The results show that the term function identification model based on Bidirectional Encoder Representations from Transformers (BERT) achieves a score of 0.90 F1. And the question-method term citation network outperforms existing keyword citation methods in revealing association patterns between scientific knowledge and improving the interpretability of the knowledge structure of the computing field. We believe that our work expands the methodology of keyword citation network and science mapping analysis and provides guidance for considering the term function in various scenarios.

Automated summary

This study proposes a term function-aware keyword citation network to address the lack of consideration for semantic roles of keywords in existing research. By identifying research questions and methods and constructing a question-method term citation network, the superiority of this network in science mapping analysis is validated. The evaluation results show that the term function identification model based on BERT achieves a score of 0.90 F1, and the question-method term citation network outperforms existing keyword citation methods in revealing association patterns between scientific knowledge and improving the interpretability of the computing field's knowledge structure. We believe that this work expands the methodology of keyword citation network and science mapping analysis and provides guidance for considering term function in various scenarios.