Intelligent recognition of high-quality academic papers: based on knowledge-based metasemantic networks

With the development of science and technology, academic papers rapidly accumulate as an essential form of research results. How to quickly discover high-quality papers among the vast amount of academic papers from fine-grained text content is of great significance in exploring the establishment of a scientific evaluation system. This paper constructs an intelligent recognition model for high-quality academic papers based on knowledge-based metasemantic networks. The knowledge elements in the ACM citation network dataset were extracted, and each knowledge element's text vector representation was obtained using SciBERT. Seven types of academic paper knowledge metasemantic networks were constructed through cosine similarity and similarity threshold. Considering the impact factor of journals and the weighted average citation metrics of papers, the metrics are used to label academic papers as high and low quality and to construct a dataset for the intelligent identification of high-quality academic papers. Mining degree centrality, mediator centrality, proximity centrality, eigenvector centrality, and clustering coefficients of knowledge elements from the perspective of knowledge metasemantic networks. Finally, the intelligent recognition model of high-quality academic papers is constructed. The experimental results show that the richer the variety of knowledge elements contained in the paper, the higher the quality of the paper. When a paper has four types of knowledge elements, the probability that the paper belongs to the high-quality category increases by 4.3% over the baseline. The research problem and solution knowledge elements contribute highly to the quality of academic papers, with the highly centered research problem knowledge element being essential. The intelligent recognition model for high-quality academic papers based on DNN has the best results, with P-value, R-value, and F1-value reaching 0.738, 0.659, and 0.696, respectively. The experimental results show that the intelligent recognition method of high-quality academic papers based on knowledge-based metasemantic networks proposed in this paper is effective.