Unsupervised topic labeling and opportunity model of social media data for enhancing automotive product design processes
Broto Widya Hartanto, Subagyo Subagyo, I Gusti Bagus Budi Dharma · Data and Information Management · 2025
This study introduces a hybrid method combining topic modeling and an opportunity model to generate novel ideas for automobile product design improvements. Furthermore, it proposes a novel unsupervised topic labeling procedure to address the limitations in current topic modeling interpretations, which are often not fully unsupervised. The procedure comprised automatic generation of labels that directly support opportunity modeling and facilitate product design development. To achieve the stated objectives, data was collected from user comments on YouTube car reviews and analyzed using various algorithms and part-of-speech rules, finding that Non-Negative Matrix Factorization with noun-adjective combinations proved most effective in generating comprehensible topic labels and capturing emotional expressions. The results revealed six underserved labels, one served right, and two overserved categories for new vehicle design improvements, providing valuable insights into user experiences . The insights provided in this context are expected to contribute to the potential improvement of vehicle attribute designs, thereby enhancing the efficiency of the entire design process.