This study aims to explore directions for improving academic supervision by analyzing graduate students' perceptions of importance and performance and comparing differences across academic disciplines. To this end, the importance and performance of 13 supervisory roles were measured using a Likert 5-point scale among 658 graduate students at a research-oriented university in the Seoul metropolitan area and analyzed through Importance-Performance Analysis (IPA). The results showed that students perceived thesis supervision and apprenticeship-based mentoring as most important but rated actual performance low, indicating a significant demand for enhanced academic guidance. By discipline, Hard Sciences showed a high demand for improving thesis and apprenticeship mentoring, while Soft Sciences emphasized the need for academic networking. These findings suggest that supervisory improvements require tailored strategies rather than universal standards. Based on this, the study proposes institutional measures: introducing formal thesis supervision courses for Hard Sciences, establishing co-supervision and networking funds for Soft Sciences, and providing research protection grants for Soft-Pure Sciences. By identifying the gap between student needs and actual experiences, this study provides a foundation for evidence-based policies to enhance the quality of graduate education.