|
|
--- |
|
|
dataset_info: |
|
|
features: |
|
|
- name: task_id |
|
|
dtype: string |
|
|
- name: sector |
|
|
dtype: string |
|
|
- name: occupation |
|
|
dtype: string |
|
|
- name: prompt |
|
|
dtype: string |
|
|
- name: reference_files |
|
|
list: string |
|
|
- name: reference_file_urls |
|
|
list: string |
|
|
- name: reference_file_hf_uris |
|
|
list: string |
|
|
splits: |
|
|
- name: train |
|
|
num_bytes: 597795 |
|
|
num_examples: 220 |
|
|
download_size: 342719 |
|
|
dataset_size: 597795 |
|
|
configs: |
|
|
- config_name: default |
|
|
data_files: |
|
|
- split: train |
|
|
path: data/train-* |
|
|
--- |
|
|
|
|
|
# Dataset for *AECBench* |
|
|
|
|
|
AECBench is a benchmark dataset for the Architecture, Engineering, and Construction (AEC) domain, built on the OpenCompass framework. It covers multiple task types ranging from basic knowledge checks to complex project-level reasoning. |
|
|
|
|
|
- 23 tasks derived from real AEC practice. |
|
|
- 4800 questions reviewed by experts. |
|
|
|
|
|
|
|
|
## Dataset Subsets |
|
|
|
|
|
### Multiple-Choice Questions (MCQ) |
|
|
|
|
|
- Subset: 1-1 规范记忆 |
|
|
Task: Memorization of building codes |
|
|
|
|
|
- Subset: 1-2 专业术语 |
|
|
Task: Recognition of technical terminology |
|
|
|
|
|
- Subset: 1-3 简称识别 |
|
|
Task: Identification of abbreviations |
|
|
|
|
|
- Subset: 2-1 规范条文解读 |
|
|
Task: Interpretation of code clauses |
|
|
|
|
|
- Subset: 2-2 设计通识问答 |
|
|
Task: General design knowledge Q&A |
|
|
|
|
|
- Subset: 2-4 规范条文解读（表格数据） |
|
|
Task: Interpretation of code clauses with tabular input |
|
|
|
|
|
- Subset: 3-1 设计决策制定 |
|
|
Task: Design decision making |
|
|
|
|
|
- Subset: 3-2 建筑类型判断 |
|
|
Task: Classification of building types |
|
|
|
|
|
- Subset: 3-3 设计决策制定（表格数据） |
|
|
Task: Design decision making with tabular input |
|
|
|
|
|
- Subset: 4-1 建筑设计中的计算 |
|
|
Task: Architectural design calculations |
|
|
|
|
|
- Subset: 4-2 MEP 设计中的计算 |
|
|
Task: MEP design calculations |
|
|
|
|
|
- Subset: 4-3 施工中的计算 |
|
|
Task: Construction-related calculations |
|
|
|
|
|
- Subset: 4-4 工程经济学中的计算 |
|
|
Task: Engineering economics calculations |
|
|
|
|
|
- Subset: 5-1-3 合规审查 |
|
|
Task: Compliance check (multiple-choice format) |
|
|
|
|
|
|
|
|
### Professional Identification |
|
|
|
|
|
- Subset: 5-1-1 专业识别 |
|
|
Task: Identify the professional field of a given text |
|
|
|
|
|
|
|
|
### Text Error Correction |
|
|
|
|
|
- Subset: 5-1-2 文本纠错 |
|
|
Task: Correct domain-specific errors in technical text |
|
|
|
|
|
|
|
|
### Information Extraction |
|
|
|
|
|
- Subset: 2-3 信息抽取 |
|
|
Task: Extract structured key information (e.g., regulation parameters, design indices) |
|
|
|
|
|
|
|
|
### Information Matching |
|
|
|
|
|
- Subset: 5-1-4 品牌合规验证 |
|
|
Task: Match product/brand information against compliance requirements |
|
|
|
|
|
|
|
|
### Proposal Recommendation |
|
|
|
|
|
- Subset: 5-3-1 设计方案推荐 |
|
|
Task: Recommend a design proposal based on project requirements |
|
|
|
|
|
|
|
|
### Report Generation |
|
|
|
|
|
- Subset: 5-3-2 报告生成 |
|
|
Task: Generate a formal AEC report (e.g., design summary, compliance report) |
|
|
|
|
|
|
|
|
### Information Consistency Check |
|
|
|
|
|
- Subset: 5-3-3 招投标信息验证 |
|
|
Task: Verify consistency of tender and bidding documents |
|
|
|
|
|
|
|
|
## Summary |
|
|
|
|
|
- Unified schema across all subsets: instruction, question, answer |
|
|
|
|
|
- Task diversity: |
|
|
Multiple-choice |
|
|
Classification |
|
|
Text correction |
|
|
Information extraction |
|
|
Compliance matching |
|
|
Proposal recommendation |
|
|
Report generation |
|
|
Consistency checking |
|
|
|
|
|
Together, these subsets cover the full spectrum of AEC domain tasks, from fundamental knowledge to applied reasoning and document validation. |
