A carregar...
Publicação
Advancing Model Transparency In Natural Language Processing: A Case Study of Explainable AI at Banco de Portugal
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 1.42 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
The efficiency of Machine Learning (ML) algorithms and projects presents a tremendous
opportunity for better decision-making in the activity of financial institutions. However, the
growing complexity of ML algorithms proves to be a considerable obstacle to the widespread
adoption of these technologies due to the lack of transparency and interpretability, deterring
users from understanding the reasoning for the decisions made by the model. Despite the
clear value and success of these ML models, uncertainty and concerns from end users intensify
this issue, forcing them to blindly trust the outcomes, especially in the context of Natural
Language Processing (NLP) projects. At Banco de Portugal, the implementation of
classification models for information requests has been very successful. However, the lack of
explainability of some results poses an obstacle to the full adoption of some models by the
business, resulting in constant skepticism about the results and fear of misinformation. We
propose a framework that systematically addresses these challenges, with the final goal of
overcoming these obstacles is a crucial step to develop trust, reduce doubt and ensure a
smooth integration of ML technologies in the decision-making practices of financial
institutions.
Descrição
Dissertation presented as the partial requirement for obtaining a Master's degree in Information Management, specialization in Knowledge Management and Business Intelligence
Palavras-chave
Explainable artificial intelligence natural language processing evaluation methods trustworthy artificial intelligence explainability SDG 9 - Industry, innovation and infrastructure SDG 12 - Responsible production and consumption SDG 17 - Partnerships for the goals