To optimize document and data analysis for Foods for Special Medical Purposes (FSMP), this study develops a multimodal framework integrating information data mining and analysis. For information mining, a dual-modality model (combining visual and geometric features) was constructed based on pdfplumber technology, addressing cross-page table fragmentation and unit omission issues. By incorporating an improved Conditional Random Field algorithm, high-accuracy reconstruction of complex document structures was achieved. For data analysis, a multidimensional evaluation framework was established, integrating econometric and nutritional metrics. Intervention analysis, entropy theory, and kernel density estimation were introduced to reveal inherent patterns in China’s special medical food market development. This methodology provides a structured database for future research and is extensible to intelligent document processing in healthcare.