PlanGlow: Personalized Study Planning with an Explainable and Controllable LLM-Driven System

TL;DR

PlanGlow addresses transparency and controllability gaps in AI-assisted self-directed learning by delivering personalized, explainable study plans with easy user control. The system combines a two-component interface, a three-step chain-of-thought generation process, Bloom's Taxonomy and ZPD-inspired reasoning, and verified video resources via the YouTube API to produce weekly and daily plans with in-line editability. In a within-subject study against GPT-4o and Khanmigo with 24 participants and expert evaluations, PlanGlow significantly improves controllability, explainability, and study-plan quality, while maintaining comparable performance and higher user preference. The findings support the practical value of explainable, controllable LLM-driven planning for self-directed learning and point to future work on diverse resources, progress monitoring, and domain-specific prompting.

Abstract

Personal development through self-directed learning is essential in today's fast-changing world, but many learners struggle to manage it effectively. While AI tools like large language models (LLMs) have the potential for personalized learning planning, they face issues such as transparency and hallucinated information. To address this, we propose PlanGlow, an LLM-based system that generates personalized, well-structured study plans with clear explanations and controllability through user-centered interactions. Through mixed methods, we surveyed 28 participants and interviewed 10 before development, followed by a within-subject experiment with 24 participants to evaluate PlanGlow's performance, usability, controllability, and explainability against two baseline systems: a GPT-4o-based system and Khan Academy's Khanmigo. Results demonstrate that PlanGlow significantly improves usability, explainability, and controllability. Additionally, two educational experts assessed and confirmed the quality of the generated study plans. These findings highlight PlanGlow's potential to enhance personalized learning and address key challenges in self-directed learning.

Figures (5)

• Figure 1: Detailed study plan of PlanGlow organizes each week into five days. (C1) explains the reasons for studying each day's topic. (C2) lists learning objectives. (C3) allows users to explore additional resources via the button connecting to (C5). (C4) displays video resources with their status. A green check icon marks 'Valid Resource', while a red icon indicates 'Invalid Resource'. (C5) displays 10 additional resources with views, likes, and video descriptions. Clicking the 'Select' button replaces the original plan's video with the selected one.
• Figure 2: The workflow of PlanGlow progresses from left to right. The system begins by collecting user inputs through an input form, in-line editing, or chat interface to create an initial plan and describe the background knowledge level. The study plan is generated through three sequential steps using the OpenAI API: Initial Generation, Critique, and Improvement. The final plan incorporates comprehensive elements, including learning objectives, content selection rationales, conceptual connections across daily and weekly units, and explanations for studying each topic. All video resources are validated and supplemented by the YouTube Data v3 API. Each block details the specific learning theories applied and parameters tuned.
• Figure 3: The interface of two systems are compared with PlanGlow in the evaluation: a GPT-4o-based system (left) and Khan Academy's Khanmigo for the 'Coach my academic and career growth' feature (right).
• Figure 4: Means and standard errors of GPT-4o-based system, Khanmigo, and PlanGlow on a 7-point Likert scale ($\ast$: $p < .05$).
• Figure 5: Means and standard errors for each evaluation assess the educational quality of generated plans on a 5-point Likert scale ($\ast\ast$: $p < .01$).