In personalized learning, facilitators need to provide learning support that appropriately navigates and scaffolds students’ learning on the basis of learning contexts, such as the content of the learning activities and the students’ learning situations. In recent years, the use of generative AI in education has increased, with studies attempting to extract students’ learning contexts from learning record data and provide learning support based on these contexts.

If a system could systematically control generative AI behaviors in learning support, the system would be able to provide appropriate learning support on the basis of students’ learning contexts in response to various learning activities. To achieve this, a system needs to be able to interpret generative AI behaviors, such as when and how to load content into generative AI, and what kind of learning support it should generate. Therefore, a formal method is needed to describe generative AI behaviors in learning support. However, no such method has yet been established.

The purpose of this study was to support learning based on students’ learning contexts using generative AI. Specifically, we introduced the Learning State Transition Diagram (LSTD) and developed a formal method of describing students’ learning activities, their learning situations, and generative AI behaviors in learning support in accordance with the LSTD.

This method enables explicit descriptions of generative AI behaviors in learning support, which can then be interpreted by the system. This will enable the system to systematically control generative AI behaviors in learning support, and the system is expected to be able to use generative AI to provide appropriate learning support based on students’ learning contexts in response to various learning activities.

We developed a formal method of describing generative AI behaviors in learning support through the following steps.
Step 1: Clarifying the framework for generative AI behaviors in learning support.
Step 2: Extracting students’ learning activities and generative AI behaviors in learning support.
We regarded the texts generated by generative AI, based on the learning context extracted from learning record data of students, as forms of learning support, and clarified the types of learning support content provided by generative AI.
Step 3: Analyzing the transition of generative AI behaviors in learning support.
We analyzed generative AI behaviors in learning support, focusing on the kinds of learning activities, the learning conditions, and the ways in which generative AI provides learning support.
Step 4: Representing generative AI behaviors in LSTD.
We represented student actions and the learning support provided by generative AI using a transition diagram that extends the theory of finite automata.
Step 5: Formally describing the LSTD.
We described the elements of the LSTD in a JSON file format so that the system can interpret the generative AI behaviors in learning support represented by the LSTD. By reading a JSON file that formally describes generative AI behaviors in learning support, the system will be able to systematically control the AI and provide students with support adapted to their learning contexts.

Verification of generative AI behaviors in learning support using the proposed method showed that the system generally guided the AI to behave appropriately.

In the future, we will evaluate its effectiveness as a type of learning support.

Keywords: Learning Support, Generative AI Behaviors, Formal Description Method, Learning State Transition Diagram, Personalized Learning.