A Study on Automatic Open-Domain Dialogue Evaluation Metrics

Abstract

Given the extensive use of conversational systems in daily life, there arises a need for an automated evaluation system tailored to such conversations. In light of this, an array of studies is underway to develop automatic evaluation metrics. Unlike conventional approaches, contemporary methodologies incorporate techniques that leverage the contextual nuances of conversation systems. In conclusion, it points out that recently explored methodologies are better suited for assessing conversations compared to classical methods; however, it is worth noting that these methodologies may also possess inherent limitations. © 2024 IEEE.