Background: Natural language processing (NLP) and machine learning technologies offer significant advantages, such as facilitating the delivery of reflective feedback in collaborative learning environments while minimising technical constraints for educators related to time and location. Recently, scholars' interest in reflective feedback has increased scientifically. However, robust empirical evidence evaluating the impacts of feedback mechanisms and innovative NLP methods on enhancing the quality of explanations in knowledge building (KB) remains limited. Objectives: This study investigated whether reflective feedback can assist primary school students in internalising and enhancing the depth of their explanations within a KB context. Method: Employing a design-based research methodology, this study involved 32 sixth-grade students from a primary school in Yangzhou, China, who engaged in a 15-week KB learning initiative. Results and Conclusions: The findings indicated that (1) students achieved significant progress in developing deep explanations, marked by improvements in logicality, consistency, convergence and structure and (2) students perceived reflective feedback as a critical factor in developing robust and profound explanations and demonstrated positively to the feedback. Implications: These findings deepened our understanding of explanatory and assessment strategies in KB and highlighted the substantial instructional, technological and practical implications of reflective feedback for educators and learners.