ScholarWorks@경상국립대학교

초록

In industry, companies conduct pre-evaluations of new product concepts to assess their potential market success and refine designs accordingly. However, conventional evaluation methods, such as focus group discussions and expert interviews, are often time-consuming and costly, limiting their scalability and frequency. With the rise of generative AI, particularly large language models (LLMs), more efficient approaches to concept evaluation have emerged. Despite these advancements, existing methods largely overlook the diverse characteristics of consumers and the need for personality incorporation, reducing their effectiveness in capturing varied customer perspectives. This study proposes a personality-incorporated generative AI framework that adapts LLMs to reflect the properties of different customer segments. We employ sentiment-based customer segmentation on smartphone reviews and fine-tune LLaMA to generate synthetic evaluations aligned with real-world consumer sentiments. Experimental results demonstrate that fine-tuned models effectively capture overall sentiment trends, though challenges remain in preserving sentiment contrast and mitigating biases toward positive outputs. Additionally, segment-aware fine-tuning enhances alignment with actual customer opinions, offering a structured approach for analyzing consumer feedback in product design. By bridging LLMs and customer segmentation, this work improves AI-driven product concept evaluations, offering a scalable, data-driven approach to support informed decision-making in new product development.

키워드