A Dataset of Scientific Experimental Events in the Field of Abiotic Stress Resistance in Rice Breeding

Based on Web of Science ™ Collecting relevant literature from the PubMed Central database between 1990 and 2023. Using an enhanced LightRAG retrieval-augmented model and a modified BERT language model, the study extracted nine types of entities—such as variety, gene, protein, and environmental factor—and eight types of relationships. The dataset adopts a five-element tuple structure, incorporating temporal and disciplinary attributes, and contains 10,409 entities and 13,034 entity relationships. Events are presented in a subject–predicate–object structure and categorized into four types: attribute-related, function-related, causal, and technical events. Throughout the data construction process, standardized cleaning, semantic annotation, expert review, and multi-model evaluation were conducted to ensure data accuracy and consistency. Compared to traditional scientific data resources, this dataset emphasizes dynamic relationships and technical pathways in experimental processes. It provides a valuable foundation for applications such as rice functional gene mining, trait improvement analysis, knowledge graph construction, and intelligent question answering systems, offering significant potential for reuse and broad applicability.