GIFT

This paper examines machine learning methods to automatically generate a large number of child scenarios from a small number of parent scenarios in support of adaptive instruction conducted in virtual simulations and game-based platforms. Adaptive instructional systems (AISs) include Intelligent Tutoring Systems (ITSs), intelligent mentors, recommender systems, personal assistants, and intelligent instructional media. AISs attempt to tailor instruction for individuals and teams based on their learning needs (e.g., knowledge or skill deficiencies), goals, and preferences. This often requires much more content than current non-adaptive systems which provide one or a very limited set of training scenarios to address a given set of learning objectives. The goal of the research described in this paper is to reduce the authoring burden for developing a large number of unique and relevant training scenarios. The methodology presented also ranks the resulting scenarios with respect to a set of author-specified learning
objectives and learner/team competency in the domain of instruction. The unique contributions of this paper are tied to its hybrid machine learning approach, and consideration for both learning objectives and learner/team competency in automatically ranking generated scenarios.

Citation: Sottilare, R. (2018, September). A Hybrid Machine Learning Approach to Automated Scenario Generation (ASG) to Support Adaptive Instruction in Virtual Simulations and Games. In Proceedings of the International Defense & Homeland Security Simulation Workshop of the I3M Conference. Budapest, Hungary, September 2018.