Artificial Intelligence in Higher Education: Bibliometric Analysis and Experimental Evidence on Students Learning

Abstract

This study examines the development and educational implications of artificial intelligence (AI) in higher education through a combination of bibliometric analysis and experimental evidence. First, publications indexed in the Scopus database from 2000 to 2024 were analyzed to map the evolution of research on AI in higher education, with attention to publication trends, thematic concentrations, and influential studies. Text mining was conducted on titles and abstracts, and term frequency-inverse document frequency (TF-IDF) weighting was used to identify representative terms. K-means clustering and Latent Dirichlet Allocation (LDA) topic modeling were then applied to detect major research themes, while PageRank analysis of the citation network was used to identify publications with high structural influence in the field. Alongside the bibliometric analysis, the study conducted an experimental investigation of ChatGPT as a formative assessment tool. Student responses were submitted to ChatGPT to generate automated feedback and grades, and the outputs were examined in terms of feedback type, instructional value, grading consistency, and agreement with human evaluation. The results suggest that students receiving ChatGPT-supported assistance showed better learning performance than those without AI support. The feedback generated by ChatGPT contained corrective, explanatory, and motivational elements, indicating its potential to provide both cognitive and affective support during learning. In addition, the comparison between AI-generated grades and human-assigned grades showed a high level of alignment, with limited evidence of systematic bias.