2026-10-12T05:29:41Z https://www.scidb.cn/oai

10.57760/sciencedb.j00133.00660 2026-06-29T08:54:14Z

2026-06-29 Source Code for Research on Retrieval-Augmented Generation Method Based on Multidimensional Causal Relation Completion doi:10.57760/sciencedb.j00133.00660 en This dataset is a compilation of experimental data and processing results for the paper titled "Research on Retrieval-Augmented Generation Method Based on Multidimensional Causal Relation Completion". It is primarily used to support the construction, retrieval, validation, and performance evaluation of the MCC-RAG method. The data is derived from public question-answering benchmark datasets—CosmosQA, MedQA, MedMCQA, and AdversarialQA—and is systematically organized by integrating causal relation extraction results, retrieval results, evaluation results, and relevant intermediate files generated during the experimental process.The data processing workflow mainly includes raw sample cleaning, field standardization, question-context-answer structure conversion, candidate causal relation extraction, event granularity normalization, temporal order verification, causal strength assessment, and experimental result logging. All relevant experiments were conducted in a Python environment, utilizing tools and frameworks for information retrieval, natural language processing (NLP), and deep learning.This dataset is mainly targeted at research in causal reasoning and retrieval-augmented generation (RAG). It does not contain specialized spatial resolution information; temporal information is only reflected in the judgment of causal temporal relations. Individual samples may contain annotation noise from public datasets themselves or minor errors introduced during model processing, but these do not affect the overall experimental analysis and result validation. Retrieval-Augmented Generation; Multi-Dimensional Causal Completion; Hallucination Suppression Zhang Qiang Zhou Hong RESTRICTED dataset Science Data Bank