LLM Technologies Complex to Reduce Turnover through Improving Workplace Communication

Abstract

Many companies have suffered from turnover of key workers due to concerns about human relationships in their workplaces. They should enhance relationships that promote all workers to engage in their work pleasantly. The issue can be solved with active listening, where a manager tries to get what a worker feels in the workspace as a listener. However, workers seldom talk frankly. The listener needs to attain listening skills through many experiences. It costs a lot. The study proposes a training method to build LLMs that work as training partners for human listeners to enhance active listening skills. Data for the fine-tuning of the LLMs contains various privacy matters. The method introduces QLoRA to enable LLMs to be trained on-premises, avoiding privacy leaks. The paper discusses the characteristics of the LLMs to improve skills. It compares the emotion recognition performance of the LLMs before and after the fine-tuning to clarify the characteristics. Experimental results reveal that LLMs demonstrate improvement in emotion. It turns out the tuned LLMs have been equipped with higher emotion recognition capability for negative emotions than for positive ones. Through practices using the LLMs, inexperienced listeners acquire skills to extract what workers keep in their minds about issues they face in the workplace.