Abstract:

Adaptive learning environments (ALE) are the future in teaching and learning. The area of application determines the type and method of adaptation. In self-regulated learning, the learner is supported in his learning process, e.g., by adapting the learning content. As a support system in university education, students are individually advised and teachers are informed about gaps in understanding, for example. The effort to develop and operate these systems is high. Research on ALEs concerns general and Machine Learning (ML) aspects of computer science. Furthermore, there are the domain specific fields of study to include: didactic, psychology and pedagogy.

Historically, ALEs use predefined decisions in advanced expert systems, whereas more recent developments in ML approaches are of great interest. By interpreting data automatically and self-optimising, data-driven ML provides human-like decision making as well as understanding and generating language.

Special requirements have to be considered to successfully use ML for ALEs. Comprehensive data must be collected, stored and processed in real time; results have to be evaluated for usability by the system itself. Data scientists have to build the model to interpret the data, which has to be filtered and pre-processed for the specific application. In general, E-learning systems offer opportunities to deliver knowledge about the learner by logging usage data. ML systems can use this data not only for user interaction but also to control the adaption of the learning system.

Machine Learning estimates the current state and predicts future states of the learner or a group of learners. Based on this knowledge, various supporting tasks are derived: Recommendation or selection of learning content; advice to the teacher on the learner's progress, tutoring of the learner by assessment or by means of feedback, etc. This feedback can be given in a straightforward manner or can be implemented in a conversation with the learner. Chatbots are ideal for this purpose, as they can additionally create an atmosphere of personal support. Contributions of domain-specific research developed should be considered, e.g., a classification of the learner based on the theory of learning progression.

This paper extracts and elaborates the challenges and perspectives from current research. The Prerequisites of proposed Learning Systems like settings, system environment, data-corpus, learning style and communication channel are listed. Methods to design and implement the System are analysed, especially the ML-Methods to build a data-driven Model. The results of the fulfilled proof of concepts and further evaluation strategies are considered. The scope of the survey is limited to recent research applying chatbot systems for adaptation.

Finally, the correlations between requirements and effort are identified. A guideline for integration and development of data-driven Adaption is elaborated, consisting of two perspectives: The data-driven chatbot as a supplement to a system: the potential features of a data-driven bot system based on existing data sources. The chatbot as a step forward in teaching: effort and benefit are compared, prerequisites are clarified.

Keywords:

Chatbot, data-driven, machine learning, tutor.