User-generated content (UGC) offers a rich source of insights into user requirements that can inform iterative product design, drive design optimization, and improve overall usability. However, implicit user requirements are often hidden within ambiguous and unstructured textual expressions, posing challenges for conventional identification methods. To address this issue, this study proposes a novel conceptual framework that effectively translates large-scale, unstructured UGC into actionable inputs for iterative product design. Specifically, it leverages DeBERTa’s disentangled attention mechanism to resolve semantic ambiguity in user requirements classification. Subsequently, it employs a bigram-enhanced topic model (BTM) to overcome data sparsity and extract coherent, implicit requirement patterns from fragmented short-text reviews. A case study involving 19,191 UGC reviews of 6 semi-automatic coffee machines demonstrates the effectiveness of the proposed framework. The user requirements classifier attains a precision of 85.9% on this dataset, which exceeded the accuracy of Bidirectional Encoder Representations from Transformers (BERT) at 84.6% and Robustly Optimized BERT Pretraining Approach (ROBERTa) at 80.9%. The topic modeling stage exhibits higher coherence than a Latent Dirichlet Allocation (LDA) baseline across categories.The proposed framework provides a validated, scalable, and generalizable pathway for converting raw UGC into structured design knowledge. The findings underscore the potential of the proposed conceptual framework to facilitate iterative design refinement, enhance user experience, and be applied across various user-centered design contexts.