In the modern society, there are a rich number of low-carbon enterprises that the explicitly/implicitly collaborated. Effectively understanding the mechanism of their complex cooperative relationships is becoming an urgent and significant problem in information processing and management. Traditionally, these cooperation behavior are analyzed in a holistic and non-quantitative way, where the complicated relationships among various enterprises cannot be well represented. In this work, we propose to understand the low-carbon entrepreneurs’ cooperation by leveraging a massive-scale dense subgraph mining technique, based on which an evolutionary graphical model is built to dynamically represent such complex relationships. More specifically, given million-scale low-carbon enterprises, we first extract multiple biologically-aware features (e.g., production value and carbon emission) to represent each of them. Based on this, a massive-scale affinity network is constructed to characterize the relationships among these enterprises. Based on this, an efficient subgraph mining algorithm (called graph shift) is deployed to discover the neighbors for each enterprise. Finally, based on the discovered neighbors of each enterprise, we can build a graphical model to represent the relationships among explicitly/implicitly-connected enterprises. The flows of multiple attributes (benefit exchange and resources swap) can be modeled effectively. To demonstrate the usefulness of our method, we manually label the attributes of 20,000 enterprises, based on which a classification model is trained by encoding the neighboring attributes of each enterprise. Comparative results have clearly demonstrated the competitiveness of our method. Moreover, visualization results can reveal the effectiveness of our method in uncovering the intrinsic distributions/correlations of million-scale enterprises.

在现代社会中，有许多显式/隐式合作的低碳企业。有效地了解其复杂合作关系的机制已成为信息处理和管理中的紧迫而重要的问题。传统上，这些合作行为是从整体和非量化的角度进行分析的，其中无法很好地表示各种企业之间的复杂关系。在这项工作中，我们建议通过利用大规模密集的子图挖掘技术来理解低碳企业家的合作，在此基础上，构建演化图形模型来动态表示这种复杂的关系。更具体地说，对于具有数百万规模的低碳企业，我们首先提取多种具有生物学意识的功能（例如，产值和碳排放量）分别代表它们。基于此，构建了大规模的亲和力网络来表征这些企业之间的关系。基于此，部署了有效的子图挖掘算法（称为图移位）来发现每个企业的邻居。最后，根据发现的每个企业的邻居，我们可以建立一个图形模型来表示显式/隐式连接的企业之间的关系。可以有效地建模多个属性（收益交换和资源交换）的流程。为了证明该方法的有效性，我们手动标记了20,000家企业的属性，在此基础上，通过对每个企业的相邻属性进行编码来训练分类模型。比较结果清楚地证明了我们方法的竞争力。此外，可视化结果可以揭示我们的方法在揭示百万规模企业的内在分布/相关性方面的有效性。