Speaker Role Identification in Call Centre Dialogues: Leveraging Opening Sentences and Large Language Models

Abstract

This paper addresses the task of speaker role identification in call centre dialogues, focusing on distinguishing between the customer and the agent. We propose a text-based approach that utilises the identification of the agent's opening sentence as a key feature for role classification. The opening sentence is identified using a model trained through active learning. By combining this information with a large language model, we accurately classify the speaker roles. The proposed approach is evaluated on a dataset of call centre dialogues and achieves 93.61% accuracy. This work contributes to the field by providing an effective solution for speaker role identification in call centre settings, with potential applications in interaction analysis and information retrieval.