Modelling Prompting Behaviours in LLM-Mediated Task Solving

Overview

This doctoral research investigates the relationship between student prompting behaviours during Large Language Model-mediated problem-solving and learning outcomes in programming contexts. The study frames LLMs not as neutral computational tools but as integral components of intelligent interactive systems whose responses influence student confidence, verification strategies, and revision patterns. The research develops behavioural models that capture temporal and semantic dimensions of student–LLM interactions, with particular attention to how learning-related signals manifest in interaction sequences and how these patterns correlate with short-term learning gains.

Methods and approach

The investigation employs computational modelling of student prompting behaviour during LLM-assisted programming tasks. The approach involves temporal and semantic analysis of interaction sequences to identify behavioural patterns and their relationships to learning outcomes. Detected patterns inform the development of automated analytical systems that generate targeted prompting recommendations to support reflection and adaptive feedback. The research design incorporates progressively scaled empirical studies to validate the behavioural models and assess the efficacy of evidence-based prompting recommendations in enhancing learning effectiveness.

Key Findings

The research project is presented as an ongoing doctoral investigation rather than a completed study with finalised results. The work describes the conceptual framework and methodological approach for modelling prompting behaviours and their relationship to learning gains, alongside the development of automated recommendation systems. The progression toward larger-scale studies suggests that comprehensive empirical findings regarding the strength and consistency of behaviour–outcome relationships remain under development.

Implications

The research contributes to understanding how user modelling can enhance the pedagogical efficacy of LLM-mediated learning environments by making implicit learning dynamics visible through interaction analysis. By identifying and characterising prompting patterns associated with learning gains, the work provides a foundation for designing adaptive feedback mechanisms that encourage more reflective and intentional LLM use among students. The framing of LLMs as interactive system components rather than neutral tools has broader implications for educational interface design, suggesting that LLM response characteristics warrant systematic consideration in learning analytics and adaptive instructional systems. Evidence-based prompting recommendations derived from student behaviour models offer potential for supporting metacognitive development and promoting more effective engagement with generative AI tools in academic contexts.

Scope and limitations

This summary is based on the study abstract and available metadata. It does not include a full analysis of the complete paper, supplementary materials, or underlying datasets unless explicitly stated. Findings should be interpreted in the context of the original publication.