A carregar...
Publicação
Explainable AI para Confiança e Transparência
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 3.84 MB | Adobe PDF |
Resumo(s)
Esta dissertação explora o desenvolvimento de sistemas de recomendação no âmbito do
projeto EARS, que combina Federated Learning e Explainable AI, tentando tanto promover a
privacidade e a transparência. O trabalho aborda desafios como a falta de explicabilidade nos
sistemas tradicionais. Foi realizada uma análise abrangente das técnicas existentes para
melhorar a transparência dos sistemas tradicionais. Este estudo destaca como a integração de
XAI pode fornecer explicações claras sobre as recomendações, aumentando a confiança dos
utilizadores.
Neste estudo foi implementado um protótipo de sistema de recomendação baseado num
modelo de Random Forest, ao qual foi aplicado métodos de Explainable AI, nomeadamente o
LIME e o SHAP, para a geração de explicações juntamente com as recomendações feitas pelo
modelo.
Os resultados obtidos demonstram como a integração de técnicas de Explainable AI pode
contribuir para a melhor compreensão das recomendações pelos utilizadores, permitindo
identificar variáveis mais influentes nas sugestões. Estes resultados mostraram como cada
ferramenta se adapta às especificidades do domínio de e-commerce.
Assim, este trabalho reforça o potencial ganho em termos de transparência e confiança com a
integração de técnicas do XAI.
This dissertation explores the development of recommendation systems based on the EARS project, leveraging Federated Learning and Explainable Artificial Intelligence techniques, aiming to promote privacy and transparency. The work addresses challenges such as the lack of explainability in traditional systems. A comprehensive analysis of existing techniques was carried out to improve the transparency of such systems. This study highlights how the integration of XAI can provide explanations for recommendations, with the main goal of increasing user trust. In this study, a prototype recommendation system was implemented based on a Random Forest model, to which Explainable AI methods, LIME and SHAP, were applied for generating explanations alongside the recommendations produced by the model. The results obtained demonstrate how the integration of Explainable AI techniques can contribute to a better understanding of recommendations by users, allowing them to identify the most influential variables in the suggestions. These findings also showed how each tool adapts to the specific circumstances of the e-commerce domain. Thus, this work reinforces the potential benefits in terms of transparency and trust gained through the integration of XAI techniques.
This dissertation explores the development of recommendation systems based on the EARS project, leveraging Federated Learning and Explainable Artificial Intelligence techniques, aiming to promote privacy and transparency. The work addresses challenges such as the lack of explainability in traditional systems. A comprehensive analysis of existing techniques was carried out to improve the transparency of such systems. This study highlights how the integration of XAI can provide explanations for recommendations, with the main goal of increasing user trust. In this study, a prototype recommendation system was implemented based on a Random Forest model, to which Explainable AI methods, LIME and SHAP, were applied for generating explanations alongside the recommendations produced by the model. The results obtained demonstrate how the integration of Explainable AI techniques can contribute to a better understanding of recommendations by users, allowing them to identify the most influential variables in the suggestions. These findings also showed how each tool adapts to the specific circumstances of the e-commerce domain. Thus, this work reinforces the potential benefits in terms of transparency and trust gained through the integration of XAI techniques.
Descrição
Palavras-chave
Recommendation Systems Explainable Artificial Intelligence Federated Learning Privacy Transparency Trust Sistemas de Recomendação Privacidade Transparência Confiança