Research into virtual role-based learning has progressed over the past decade. Modern issues include gauging the difficulty of designing a goal system capable of meeting the requirements of students with different knowledge levels, and the reasonability and possibility of taking advantage of the well-designed formula and techniques served in other research fields to improve role-based tutoring. In this paper, we attempt to develop a comprehensive and adaptable goal system and intelligent tutors in an educational bioscience game by proposing a hybrid approach. Our solution supports multi-user collaborations and competitive play, and integrates a data mining model to help discover student play patterns. The overarching aim is to make informed tutoring decisions, and improve student learning and efficiency, as they work through each module in the game. [For the full proceedings, see ED562107.]