Scholarlink

<p>Personality influences how individuals respond to situations and interact within teams, making its prediction valuable in domains such as recruitment, counselling, and military operations. The Myers-Briggs Type Indicator (MBTI) is widely used to assess personality, but traditional assessments require expert supervision and are time-consuming. With the growing presence of social media, this study explores automatic MBTI personality prediction using text data from online interactions. We address key challenges in MBTI classification, including data imbalance and low classification accuracy, by introducing a dual-tier oversampling strategy that combines GPT–2–based contextual data generation with the Synthetic Minority Oversampling Technique (SMOTE). Various word embedding methods and machine learning classifiers were evaluated, with ensemble learning techniques (voting, stacking, and blending) yielding the best performance. The proposed approach achieved an average accuracy of 90.39% and an F1-score of 0.9037, outperforming existing models. This study demonstrates the effectiveness of combining contextual augmentation with ensemble learning, offering a robust framework for scalable and reliable MBTI personality prediction from online text data.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

MBTI personality prediction using GPT-2 LLM augmentation and ensemble machine learning approaches

  • 期刊论文

  • 发表日期：

    2026年2月1日

查看全文
  • Devraj Patel,
  • Sunita V. Dhavale

摘要

Personality influences how individuals respond to situations and interact within teams, making its prediction valuable in domains such as recruitment, counselling, and military operations. The Myers-Briggs Type Indicator (MBTI) is widely used to assess personality, but traditional assessments require expert supervision and are time-consuming. With the growing presence of social media, this study explores automatic MBTI personality prediction using text data from online interactions. We address key challenges in MBTI classification, including data imbalance and low classification accuracy, by introducing a dual-tier oversampling strategy that combines GPT–2–based contextual data generation with the Synthetic Minority Oversampling Technique (SMOTE). Various word embedding methods and machine learning classifiers were evaluated, with ensemble learning techniques (voting, stacking, and blending) yielding the best performance. The proposed approach achieved an average accuracy of 90.39% and an F1-score of 0.9037, outperforming existing models. This study demonstrates the effectiveness of combining contextual augmentation with ensemble learning, offering a robust framework for scalable and reliable MBTI personality prediction from online text data.