This is a silver corpus for Plant-based Food Fermentation Information Extraction task. This dataset focuses on plant-based food fermentation and comprises 2,500 abstracts retrieved from PubMed. The data was automatically annotated using a Large Language Model (LLM), resulting in a structured JSON format that includes a total of 23,563 entities. These entities are categorized into four distinct classes to capture the complexity of the fermentation process: Molecule (8,809 entities), Microbe (7,677 entities), Plant/Food (5,151 entities), and Habitat (1,926 entities). By mapping the relationships between microbial strains, raw plant substrates, and resulting aromatic or chemical compounds, this dataset provides a comprehensive resource for knowledge graph construction and relation extraction within the domain of sustainable food science.