Information fusion in consumer complaints: A LLM-Guided topic model for product quality improvement

The rise of user-generated content (UGC) on e-commerce platforms provides rich and timely information for product quality improvement. User reviews and ratings from UGC are key sources for analyzing consumers’ perceptions of product and service quality. However, the information fusion of reviews and ratings in consumer complaints has not been explored deeply. Furthermore, interpreting the outputs of topic models in a meaningful and actionable way remains a significant research challenge. This paper proposes the LLM-Guided Rating-Aware Neural Topic Model (LLM-Guided RANTM), a novel framework that fuses online reviews and ratings under a neural topic modeling architecture. The model incorporates a self-attention mechanism to identify multi-rooted hierarchical topic structures, while leveraging large language model (LLM) to enhance topic coherence and interpretability. A case study of China’s new energy vehicle (NEV) market demonstrates the effectiveness of the proposed model. Compared with several state-of-the-art baselines, the proposed model shows significant improvements in topic quality across multiple quantitative metrics. Ablation experiments further confirm that the model benefits from both the information fusion and the guidance of LLM. Combined with importance–performance analysis (IPA), these results provide valuable insights for product improvement and decision-making for stakeholders including manufacturers, product managers, and service designers.