Business-to-business software systems are typically highly specialized and functionally complex, making it essential to generate reliable personas that accurately reflect end-user requirements during development. Moreover, because user needs evolve over time, it is critical to establish a data-driven mechanism that continuously updates personas and promptly reflects these changes in development to ensure sustained value delivery. Traditional persona generation methods rely on clustering qualitative and quantitative data, a process that requires specialized expertise and incurs significant time and cost. This paper proposes an automated, continuous persona generation method that identifies user requirements from continuously collected data, eliminating the need for extensive manual research. The proposed approach enables rapid and up-to-date understanding of user needs by leveraging large language models to analyze qualitative data and dynamically construct and update knowledge graphs. We applied this method to a case study using inquiry call logs from a customer support center. The results show that the proposed method outperformed a clustering-based baseline in approximately 86% of cases in a question-answering task used to assess the structuring and retrieval of user requirements. Additionally, the generated personas effectively clarified user challenges and improvement areas, underscoring the value of continuous, data-driven persona generation.