"Dataset of Experiment 1 for KGAICQA"

"To achieve the experimental 1 goal, we constructed a dataset comprising 123 questions, each associated with 10 KG entries retrieved via cosine and Euclidean similarity calculations. Using a curriculum-theme coverage strategy, all questions were manually aligned with the specific knowledge points in the Guangzhou K\u201312 AI textbooks. This construction encompasses the core knowledge distributed across 8 instructional units and 12 lessons. Two educational technology experts specializing in AI education manually verified the validity of each question in terms of content validity, expression clarity, thematic relevance, and pedagogical relevance. The reliability and validity of automated evaluation depend on the scale sufficiency and diversity of the datasets. To address this requirement, the original panel of subject-matter experts was reconvened to expand the dataset from Experiment 1, and Guangzhou K\u201312 AI textbooks were used as the foundational source. This process yielded a final corpus of 1,098 questions that comprehensively covered the curricular knowledge points. Following Bloom\u2019s revised taxonomy[65], the questions were classified into three levels of cognitive difficulty: simple (n = 364), moderate (n = 367), and difficult (n = 367). The descriptive statistics of this experimental dataset are summarized in TABLE IV. Upon establishing the validity of the compiled problem set, further data expansion was halted. This decision was driven by high computational costs\u2014totaling over 9,226,652 tokens\u2014as well as predefined experimental constraints and allocated computational resource quotas.  "