RB-ELSAD is a high-confidence dataset of semantic associations between entities and literature, designed for rice breeding. By integrating domain entities (such as genes, proteins, and traits) with massive scientific literature, and utilizing a multi-pathway hybrid retrieval and supervised discrimination framework, this dataset constructs high-confidence semantic associations between the two, totaling 509,788 semantic relations. The dataset aims to break the data silos between domain resources and scientific literature, providing critical data infrastructure for research in rice gene mining, trait dissection, and intelligent breeding decision-making. This dataset contains two core components:(1) Domain Layer: Breeding entities and their biological relations.(2) Resource Layer: Scientific literature and its semantic associations with domain entities.