<![CDATA[Composition labels on consumer products serve as an essential source of information for consumers in making purchasing decisions and for producers in ensuring regulatory compliance. However, manual reading of labels is prone to errors due to small font sizes, blurred images, low resolution, and perspective distortion, which reduce the accuracy of information recognition. This issue becomes increasingly critical as many label compositions contain ingredients associated with metabolic diseases, such as added sugar, artificial sweeteners, and carbohydrates, whose risks are well-documented in epidemiological literature. While numerous systematic reviews have addressed OCR and NLP in document intelligence, no prior review has systematically mapped the integration of OCR robustness, BERT-based semantic extraction, and food label-specific challenges in the context of multilingual Indonesian consumer products. To address this gap, this study conducts a Systematic Literature Review (SLR) following PRISMA guidelines, synthesizing recent advances in Transformer-based OCR and BERT-based NLP for consumer product label analysis. The review aims to map OCR performance under real-world conditions, identify best practices in preprocessing and post-correction, and evaluate end-to-end integration with BERT for semantic understanding of label compositions. The expected outcome is a theoretical contribution in the form of an integrative OCR–BERT framework proposal, along with practical design recommendations for future systems aimed at supporting nutritional literacy and informed consumer decision-making. Empirical validation of the proposed framework remains a direction for future research.]]>
