Percorrer por autor "DOMINGUES, MIGUEL RODRIGUES"
A mostrar 1 - 1 de 1
Resultados por página
Opções de ordenação
- Development of Intelligent Mechanisms for Contextual Support of Writing and Convergence of Proposals in Participatory ProcessesPublication . DOMINGUES, MIGUEL RODRIGUES; Marreiros, Maria Goreti CarvalhoDigital participatory processes allow citizens to contribute to public decision-making through online platforms. However, many proposals are submitted with limited clarity, incomplete information, or weak alignment with process rules. In addition, similar or complementary ideas are often submitted separately, increasing fragmentation and reducing opportunities for collaboration. This dissertation addresses these challenges by designing, implementing, and evaluating a proof of concept for AI-assisted proposal writing and semantic convergence in digital participatory processes. The work was developed within the CoParticipation project and aligned with the EMPATIA framework. The prototype combines a Laravel and Livewire web platform with a FastAPI-based AI service responsible for regulatory validation, proposal improvement, document ingestion, retrieval, semantic convergence, and local language model interaction. The solution uses Natural Language Processing, Large Language Models, Retrieval-Augmented Generation, embeddings, vector retrieval, and LoRA-based model adaptation. Its main functionalities are a two-stage proposal review workflow and a semantic convergence mechanism that identifies similar or complementary proposals before submission. The evaluation combined automatic tests, non‑functional quality assessment, and expert feedback. Results show that contextual knowledge and model adaptation improve regulatory validation and proposal review, while the prototype meets the defined quality requirements. Experts confirmed the usefulness and adoption potential of the solution. Overall, the approach enhances proposal clarity, supports regulatory awareness, identifies related proposals, and provides effective contextual assistance without compromising citizen control or administrative responsibility.