Transforming multimodal online reviews into competitor-aware product improvement decisions: a large language model-assisted framework

Online customer reviews provide rich evidence for product iteration, but their fragmented, multimodal, and competitive nature makes it difficult to convert user feedback into actionable priorities. This study proposes CompReview, a large language model-assisted framework for multimodal review mining and competitor-aware product improvement. CompReview extracts aspect-level features, opinions, sentiment scores, and supporting evidence through a structured prompting schema, normalizes fragmented expressions into a fixed feature taxonomy, and extends importance-performance analysis with competitor-relative gaps. The framework is evaluated on 156,789 Apple and Huawei smartphone reviews. A leakage-safe 70/15/15 split of the 1200 annotated JD.com reviews yields F1-scores of 0.920 for feature extraction, 0.930 for sentiment classification, and 0.900 for attribute normalization. On an independent 400-review Amazon gold set, the corresponding scores are 0.870, 0.890, and 0.840. Five fixed-configuration runs show low variability (SDs of 0.006, 0.007, and 0.011), while the main quadrant agreement is 0.870. Apple’s strengths concentrate in appearance and packaging, whereas price, after-sales service, logistics, battery life, and selected experience attributes remain improvement priorities. Ignoring competitor performance misclassifies 33.3% of high-priority attributes. CompReview therefore provides a reproducible review-to-decision pipeline for product redesign, service optimization, and market positioning.