We use cookies to ensure that we give you the best experience on our website. By continuing to browse this repository, you give consent for essential cookies to be used. You can read more about our Privacy and Cookie Policy.

Durham e-Theses
You are in:

Deep Reinforcement Learning Approaches for Technology Enhanced Learning

LI, ZHAOXING (2023) Deep Reinforcement Learning Approaches for Technology Enhanced Learning. Doctoral thesis, Durham University.

PDF (Zhaoxing Li' s PhD Thesis) - Accepted Version


Artificial Intelligence (AI) has advanced significantly in recent years, transforming various industries and domains. Its ability to extract patterns and insights from large volumes of data has revolutionised areas such as image recognition, natural language processing, and autonomous systems. As AI systems become increasingly integrated into daily human life, there is a growing need for meaningful collaboration and mutual engagement between humans and AI, known as Human-AI Collaboration. This collaboration involves combining AI with human workflows to achieve shared objectives.

In the current educational landscape, the integration of AI methods in Technology Enhanced Learning (TEL) has become crucial for providing high-quality education and facilitating lifelong learning. Human-AI Collaboration also plays a vital role in the field of Technology Enhanced Learning (TEL), particularly in Intelligent Tutoring Systems (ITS). The COVID-19 pandemic has further emphasised the need for effective educational technologies to support remote learning and bridge the gap between traditional classrooms and online platforms. To maximise the performance of ITS while minimising the input and interaction required from students, it is essential to design collaborative systems that effectively leverage the capabilities of AI and foster effective collaboration between students and ITS.

However, there are several challenges that need to be addressed in this context. One challenge is the lack of clear guidance on designing and building user-friendly systems that facilitate collaboration between humans and AI. This challenge is relevant not only to education researchers but also to Human-Computer Interaction (HCI) researchers and developers. Another challenge is the scarcity of interaction data in the early stages of ITS development, which hampers the accurate modelling of students' knowledge states and learning trajectories, known as the cold start problem. Moreover, the effectiveness of Intelligent Tutoring Systems (ITS) in delivering personalised instruction is hindered by the limitations of existing Knowledge Tracing (KT) models, which often struggle to provide accurate predictions. Therefore, addressing these challenges is crucial for enhancing the collaborative process between humans and AI in the development of ITS.

This thesis aims to address these challenges and improve the collaborative process between students and ITS in TEL. It proposes innovative approaches to generate simulated student behavioural data and enhance the performance of KT models. The thesis starts with a comprehensive survey of human-AI collaborative systems, identifying key challenges and opportunities. It then presents a structured framework for the student-ITS collaborative process, providing insights into designing user-friendly and efficient systems.

To overcome the challenge of data scarcity in ITS development, the thesis proposes two student modelling approaches: Sim-GAIL and SimStu. SimStu leverages a deep learning method, the Decision Transformer, to simulate student interactions and enhance ITS training. Sim-GAIL utilises a reinforcement learning method, Generative Adversarial Imitation Learning (GAIL), to generate high-fidelity and diverse simulated student behavioural data, addressing the cold start problem in ITS training.

Furthermore, the thesis focuses on improving the performance of KT models. It introduces the MLFBKT model, which integrates multiple features and mines latent relations in student interaction data, aiming to improve the accuracy and efficiency of KT models. Additionally, the thesis proposes the LBKT model, which combines the strengths of the BERT model and LSTM to process long sequence data in KT models effectively.

Overall, this thesis contributes to the field of Human-AI collaboration in TEL by addressing key challenges and proposing innovative approaches to enhance ITS training and KT model performance. The findings have the potential to improve the learning experiences and outcomes of students in educational settings.

Item Type:Thesis (Doctoral)
Award:Doctor of Philosophy
Keywords:Human-AI Collaboration, Technology Enhanced Learning, Intelligent Tutoring Systems, Knowledge Tracing, Decision Transformer, Generative Adversarial Imitation Learning, MLFBKT, LBKT.
Faculty and Department:Faculty of Science > Computer Science, Department of
Thesis Date:2023
Copyright:Copyright of this thesis is held by the author
Deposited On:02 Nov 2023 14:47

Social bookmarking: del.icio.usConnoteaBibSonomyCiteULikeFacebookTwitter