This thesis addresses the adoption issue of learning analytics dashboards (LAD) and it proposes an adaptation-centered approach to improve their use and their impact on learning.

A systematic literature review covering publications from 2017 to 2024, conducted following the PRISMA methodology, was carried out to identify existing approaches to LAD adaptation, as well as how these concepts are perceived within the scientific community. Building on this analysis, an empirical study based on a questionnaire was conducted with twenty French-speaking experts in the field of technology-enhanced learning. This study examined the relevance and importance of input variables and output dimensions involved in LAD adaptation.

Based on these findings, the thesis proposes an adaptation model for LADs that aims to articulate key adaptation elements, including input data, constraints, and adaptation knowledge, in order to support adaptation decision-making and generate LAD adaptation proposals. The model also adopts a broader integration perspective by facilitating the reuse of existing methods, processes, and knowledge within the community

Finally, a preliminary feasibility study was conducted through an implementation based on a retrieval-augmented generation (RAG) approach using large language models (LLMs). This experiment aimed to evaluate the system’s ability to produce adaptation decisions consistent with a generic representation of LADs, the learning context, and the learner profile. The results highlight the potential of this approach and open new research perspectives for the development of LAD adaptation systems.