Personalized Learning Path Recommendation using Graph Reinforcement Learning

A learning path is a set of sequentially accessed learning resources or objects that help learners progress in their learning journey. Researchers have introduced a range of methodologies for personalizing learning paths in the last decade, employing various computational techniques and approaches. Advanced computational techniques like supervised and un-supervised machine learning algorithms are often used in predicting efficient and personalized learning paths. In this work, we present a learning path recommended system based on the learner’s current knowledge state using graph reinforcement learning. The start point and end point of the learning path are identified from the learner’s query and knowledge base. Then, a reinforcement learning algorithm is used over a search space represented by a corpus. The corpus is composed of learning objects and their interconnections, represented as a graph where each node is a learning object, and each edge represents the degree of relationship of two joining nodes. We employ the Q-learning algorithm to recommend an appropriate learning path for the user’s query that maximizes the use of their existing knowledge. An experimental study is performed on the Wikipedia corpus to discover different learning paths. The initial results are promising, which indicates the efficiency of the proposed methodology.