Assessment and Improvement of Customer Service Speech with Multiple Large Language Models
So Watanabe, Chee Siang Leow, Junichi Hoshino, Takehito Utsuro, Hiromitsu Nishizaki · 2024
This paper introduces a framework using multiple large language models (LLMs) to assess and enhance the customer service interactions of a staff in service industry. Effective communication with customers is pivotal for better customer satisfaction. To enhance these skills, precise and constructive feedback is crucial for customer service staff. This study employs multiple LLMs within a round-table discussion framework, named "ReConcile" to evaluate and suggest improvements for customer service dialogues. Proposed method scores a customer service speech and suggests suitable response. An subjective experiment was conducted where human subjects compared the effectiveness of customer service speech assessments and response suggestions generated by both a single LLM and the ReConcile method. Results showed that the scores with ReConcile were closer to human senses compared to a single LLM which indicate the suggestions for improving the customer service staff’s speech.