Personalized Learning Path Planning with Goal-Driven Learner State Modeling

TL;DR

Pxplore introduces a goal-driven, reinforcement-based PLPP framework that jointly optimizes long-horizon learning paths and dynamic learner state representations. By encoding structured learner states and an automated abstract reward, it enables GRPO-based policy refinement atop SFT initialization, and couples this with an LLM-driven architecture for real-time profiling, planning, and adaptive delivery. Extensive experiments and a real-world user study demonstrate superior pedagogical alignment, planning quality, and learning gains compared with baselines, validating both the training paradigm and the end-to-end system. The work advances practical, scalable, goal-aligned personalized learning with open-source resources to foster future research.

Abstract

Personalized Learning Path Planning (PLPP) aims to design adaptive learning paths that align with individual goals. While large language models (LLMs) show potential in personalizing learning experiences, existing approaches often lack mechanisms for goal-aligned planning. We introduce Pxplore, a novel framework for PLPP that integrates a reinforcement-based training paradigm and an LLM-driven educational architecture. We design a structured learner state model and an automated reward function that transforms abstract objectives into computable signals. We train the policy combining supervised fine-tuning (SFT) and Group Relative Policy Optimization (GRPO), and deploy it within a real-world learning platform. Extensive experiments validate Pxplore's effectiveness in producing coherent, personalized, and goal-driven learning paths. We release our code and dataset to facilitate future research.

Figures (10)

• Figure 1: Comparison of traditional and LLM-based learning path planning. Traditional methods produce static paths, while LLMs struggle to capture long-term progress.
• Figure 2: Overview of the Pxplore framework, including: (1) an Reinforcement-based Training Paradigm for goal-aligned learning path planning; and (2) an LLM-driven educational architecture for adaptive profiling, planning and delivery.
• Figure 3: Example of Pxplore's pre-planning architecture that generates a structured learner profile.
• Figure 4: Example of Pxplore's post-planning architecture that enables adaptive delivery of selected action.
• Figure 5: Heatmap showing the improvement in evaluation scores on a 5-point Likert scale across pedagogical dimensions, calculated as $\text{Score}_{{\textit{Pxplore}}\xspace} - \text{Score}_{\text{Baseline}}$.