院校就业去向数据集:
gaokao/real_data/uni_employment.json。来自各校毕业生就业质量报告,权威来源受限蒸馏,缺失即留空、绝不编造。The uni employment dataset:
gaokao/real_data/uni_employment.json. Distilled from each school’s 就业质量报告 under an authoritative-source restriction; honest-null where unpublished, never fabricated.
覆盖 / Coverage
中文。 共 2347 所院校;有数据 2140 所(其余无公开报告 → no_report:true)。按字段:就业率 1547 · 升学率 1547 · 出国境率 1547 · 起薪 2140。起薪最稀疏——多数报告不披露,且通常是全校/学院级而非单一专业。
English. 2347 schools; 2140 with data (the rest publish no report → no_report:true). By field: 就业率 1547 · 升学率 1547 · 出国境率 1547 · salary 2140. Salary is the sparse one — most reports omit it, and it’s usually whole-school / college-level, not per-specific-major.
结构 / Schema
{ "uni": "中国农业大学", "report_year": "2024",
"destinations": {"升学率": 58.13, "出国境率": 12.72, "就业率": 94.71, "考公考编": null},
"salary": {"应届平均月薪_rmb": 10360, "level": "全校级", "note": "2024届本科毕业生平均月薪"},
"top_industries": ["政府及公共管理","教育业","农林牧渔业","金融业"],
"top_employers": ["国家电网","银行","烟草局"],
"by_college": [], "sources": ["中国农业大学本科教学质量报告(2023–2024)"],
"no_report": false, "provenance": "web-distilled 2026-06-16 (deepseek-v3 + web); authoritative-source restricted" }怎么被用 / How it’s used
中文。 由 grounding.py 的 lookup_uni() + uni_employment_block() 接进画像:按选项标签里的学校名(最长匹配)取出该校的 就业率/升学率/出国率/起薪(带口径)+ 行业/雇主 + 来源,注入调查员上下文;无数据则整块省略。它替代了早先被标记为 miscalibrated 的”专业级蒸馏薪资”。机制见 grounding-layer。
English. Wired into the profile by lookup_uni() + uni_employment_block() in grounding.py: it pulls the school’s 就业率/升学率/出国率/salary (with 口径) + industries/employers + source by longest-name match on the option label, injected into the investigator context; omitted entirely when absent. It replaced the earlier miscalibrated “distilled per-major salary.” Mechanism: grounding-layer.
Related
data-overview · data-schools-majors · grounding-layer · yulai-integration