Integrating Natural Language Models with Bayesian Networks for Explainable Machine Learning

Abstract

The advancements brought by machine learning algorithms have significantly contributed to society, notably natural language models based on transformers, which have the capability to interpret user requests and respond in natural language. However, the lack of explainability in these kind models has been a barrier to their use in critical sectors. As an alternative, probabilistic models such as Bayesian networks have gained prominence for their ability to provide explainability and to be updated with specialist knowledge, but have limited adoption between professionals without a background in statistics caused by their mathematical complexity. This research aims to integrate these two approaches, resulting in a tool that will enable users to request and comprehend inferences in probabilistic models in an accessible manner through natural language.