AI-DRIVEN LEARNING ANALYTICS AND THEIR EFFECT ON PERSONALIZED FEEDBACK IN BLENDED HIGHER EDUCATION COURSES AT NANJING NORMAL UNIVERSITY, CHINA

Abstract

The present paper discusses the influence of AI-guided learning analytics on personalized feedback in a large-scale blended course in English as a Foreign Language (EFL) at Nanjing Normal University in China based on the aspects of feedback design, literacy, and instructor mediation. The study combines PLS-SEM with bootstrapping, LMS log analysis, validated surveys, and semi-structured interviews (employed as a qualitative subsample of 28 students, 7 instructors; N = 400 students). Results indicate that timely and focused AI-generated feedback positively but not directly affects behavioral engagement ( = .38, p =.001) but does not affect academic performance ( =.07, p =.21). The interaction is a mediator of the improved outcome (indirect effect =.19, 95 percent interval = [.11, .28]). This relationship is mediated by feedback literacy: high-literacy students demonstrate significant gains ( =.52, p <.001), whereas low-literacy students need mediation by the instructor to gain the same (interaction =.44, p <.001). Qualitative knowledge highlights that trust and actionability are based on human validation and not algorithmic accuracy. These findings are detrimental to dichotomous LA vs. no-LA methods, as it is evident that AI is effective within the humanized ecosystem, in which analytics becomes input, teachers contextualize the inputs, and students develop the interpretation ability. The research provides a resource-limited scalable, fair model of AI integration in high-enrollment EFL programs, whose findings apply in the global higher education.