Controlling or directing? Text mining to decode supervisor-graduate student relationship

This study explores the Supervisor-Graduate Student Relationship (SGSR) using text mining and introduces a 2 × 2 quadrant model based on "control level" and "direction level," identifying four relationship types: Juice Coach (High-Control, High-Direction), Shadow Hermit (Low-Control, Low-Direction), Pilot Captain (High-Direction, Low Control), and Alchemy Master (High-Control, Low-Direction). It also examines how these relationships vary across disciplines and geo-economic contexts, finding that social sciences favor Alchemy Master styles, while economically underdeveloped regions show stronger control. A dataset of 25,219 evaluation texts from 445 Chinese universities was analyzed using BERT-TextCNN, BERTopic modeling, and BERT embedding-based semantic similarity, showing strong performance in sentiment analysis and thematic clustering. The findings offer not only theoretical insights into SGSR dynamics but also concrete practical implications: the proposed framework enables institutions to systematically monitor supervisory quality, identify at-risk relationships through routine feedback, and trigger timely interventions-functioning as an early warning system to protect student well-being and support faculty development. This work thus provides both a novel typology of SGSR and a scalable, data-driven tool for improving research supervision in real-world educational settings.