当前位置: X-MOL 学术Comput. Aided Civ. Infrastruct. Eng. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Clustering of designers based on building information modeling event logs
Computer-Aided Civil and Infrastructure Engineering ( IF 8.5 ) Pub Date : 2020-04-24 , DOI: 10.1111/mice.12551
Yue Pan, Limao Zhang, Miroslaw J. Skibniewski

A network‐enabled event log mining approach is proposed for a deep understanding of the Building Information Modeling (BIM)‐based collaborative design work. It proposes a novel algorithm termed node2vec‐GMM combining a graph embedding algorithm named node2vec and a clustering method named Gaussian mixture model (GMM) to cluster designers within a network into several subgroups, and then makes cluster analysis. Its superiority lies in the efficient feature learning ability to preserve network structure and the powerful clustering ability to tackle uncertainty and visualize results, which can directly return the cluster embedding. As a case study, a directional network with 68 nodes (designers) and 436 ties (design task transmissions) is constructed based on retrieved data from 4GB real BIM event logs. The node2vec learns and projects the network feature representation into a 128‐dimensional vector, which is learned by GMM to discover three possible clusters owning 15, 26, and 27 closely linked designers. Analysis of each cluster is performed from node importance measurement and link prediction to identify information spreading and designers’ roles within clusters. Our new algorithm node2vec‐GMM is proven to better improve clustering quality than other state‐of‐the‐art methods by at least 6.0% Adjusted Rand Index and 13.4% Adjusted Mutual Information. Overall, the designer clustering process provides managers with data‐driven support in both monitoring the whole course of the BIM‐based design and making reliable decisions to increase collaboration opportunities.

中文翻译:

基于建筑信息建模事件日志的设计师聚类

提出了一种基于网络的事件日志挖掘方法,以深入了解基于建筑信息模型(BIM)的协作设计工作。它提出了一种称为node2vec‐GMM的新颖算法将名为node2vec的图形嵌入算法和名为高斯混合模型(GMM)的聚类方法相结合,将网络中的设计者分为几个子组,然后进行聚类分析。它的优势在于有效的特征学习能力可以保留网络结构,强大的聚类能力可以解决不确定性并可视化结果,从而可以直接返回聚类嵌入。作为案例研究,基于从4GB实际BIM事件日志中检索到的数据,构建了具有68个节点(设计者)和436个联系(设计任务传输)的定向网络。node2vec将网络特征表示学习并将其投影到128维向量中,GMM通过学习可以发现拥有15个,26个和27个紧密链接的设计器的三个可能的集群。从节点重要性测量和链接预测执行每个群集的分析,以识别群集中的信息传播和设计者的角色。我们的新算法经过实践证明,node2vec-GMM至少比其他现有技术方法改善了群集质量,调整后的兰德指数至少达到了6.0%,调整后的相互信息达到了13.4%。总体而言,设计师集群过程为经理提供了数据驱动的支持,既可以监视基于BIM的设计的整个过程,又可以做出可靠的决策来增加协作机会。
更新日期:2020-04-24
down
wechat
bug