Personalizing Learning and Empowering Educators through Multi-Dimensional Clustering, Teacher Dashboards, and AI-Hybrid Learning Systems

The shift from traditional classroom settings to technology-supported learning environments has led to the adoption of learning analytics and artificial intelligence (AI) in education. These technologies promise to support personalized learning by analyzing student behavior, guiding teacher decision-making, and delivering targeted feedback. However, significant challenges remain in modeling complex learning processes like self-regulated learning (SRL), designing teacher-facing tools that are both actionable and interpretable, and developing student-facing interventions that promote meaningful engagement.

This dissertation addresses these gaps through three main contributions. First, it proposes a multidimensional clustering pipeline to identify SRL profiles across diverse contexts, including Flipped Classrooms and Vocational Education settings. Second, it investigates how teacher-facing dashboards can be designed to support actionable insights through visualization techniques, storytelling elements, and multimodal analytics. Third, it explores the use of retrieval-based and generative AI systems to deliver personalized examples and reflective prompts that support procedural writing and metacognition.

We propose the following technical contributions:

• A multidimensional clustering pipeline that identifies interpretable learner profiles across behavioral dimensions using a two-step approach.
• A multimodal validation framework that assesses the alignment between indicators from two or more modalities.
• A human-in-the-loop AI task generation pipeline that produces cognitive tasks aligned with Bloom's taxonomy by integrating LLMs and LMs with expert review.
• A retrieval-based adaptive example learning pipeline that delivers personalized and relevant examples using semantic similarity, language model scoring, and domain-specific criteria.
• A hybrid-AI example learning pipeline that generates personalized examples and instructional explanations using domain-specific criteria and LLMs.

We conducted empirical evaluations across diverse educational settings and stakeholder groups including:

• A longitudinal study with 79 students over 10 weeks to assess alignment between SRL survey responses and behavioral data in an Intelligent Tutoring System called Lernnavi.
• A controlled experiment with 100 teachers to evaluate the perceived clarity, appeal, and actionability of different visualization strategies for communicating clustering results.
• A qualitative study with 19 teachers, involving semi-structured interviews, to assess the effectiveness and contextual applicability of a dashboard prototype adapted to two educational settings: Vocational Education and Flipped Classrooms.
• A controlled experiment with 128 participants to evaluate data storytelling elements in visualizations across varying levels of cognitive task types.
• A classroom study with 12 educators analyzing their own data from 36 in-the-wild classroom sessions, reflecting o