BIM-Edit
收藏资源简介:
BIM-Edit是由罗斯托克大学与克劳斯塔尔工业大学联合构建的基准数据集,旨在评估大语言模型在建筑信息模型自然语言编辑任务中的性能。该数据集包含324项编辑任务,覆盖11个真实建筑模型与36个合成场景,数据以行业基础类格式存储,平均每个复杂模型包含614.88个元素与2088.88组关系。数据集通过专家手动构建,采用模块化任务语法设计,涵盖创建、更新、删除三大操作类型,并区分直接、空间、拓扑三种指令变体。该数据集主要应用于建筑信息建模领域,致力于解决大语言模型在结构化工程设计工作流中存在的场景理解不足、语义拓扑一致性维护困难等核心问题。
BIM-Edit is a benchmark dataset jointly constructed by the University of Rostock and Clausthal University of Technology, aiming to evaluate the performance of Large Language Models (LLMs) in natural language editing tasks for Building Information Modeling (BIM). This dataset includes 324 editing tasks, covering 11 real-world building models and 36 synthetic scenarios, with data stored in industry-standard basic formats. On average, each complex model contains 614.88 elements and 2088.88 relationship groups. The dataset was manually built by domain experts, designed with modular task grammar, covering three operation types: creation, update and deletion, and distinguishing three instruction variants: direct, spatial and topological. This dataset is mainly applied in the field of building information modeling, and is committed to solving core problems existing in LLMs in structured engineering design workflows, such as insufficient scene understanding and difficulties in maintaining semantic and topological consistency.

- 1BIM-Edit: Benchmarking Large Language Models for IFC-Based Building Information Modeling罗斯托克大学; 克劳斯塔尔工业大学 · 2026年




