The graph stream has recently arisen in many interactive scenarios. Characterizing with large volume and high dynamic, graph streams are known to be difficult for high-speed summary and analysis, especially provided with limited resource availability. Existing solutions mainly use sketch-based methods to estimate the weight of items (e.g., Count-Min Sketch) and preserve the underlying graph structure information (e.g., TCM). Unfortunately, these solutions neither support complex graph-based queries nor achieve efficient real-time queries. In view of these limitations, we design DMatrix, a novel 3-dimensional graph sketch to facilitate fast and accurate queries in graph stream. Both structural query and weight-based estimation are supported with DMatrix. Through the integration of representative key reservation and majority voting, DMatrix can effectively narrow the error bounds of queries with real-time response efficiency. Both theoretical analysis and experimental results confirm that our solution is superior in accuracy and efficiency comparing with the state-of-the-art.

图流最近出现在许多交互场景中。众所周知，以大容量和高动态为特征的图形流难以进行高速汇总和分析，尤其是在资源可用性有限的情况下。现有的解决方案主要使用基于草图的方法来估计项目的权重（例如，Count-Min Sketch）并保留底层的图结构信息（例如，TCM）。不幸的是，这些解决方案既不支持复杂的基于图的查询，也无法实现高效的实时查询。鉴于这些限制，我们设计了 DMatrix，一种新颖的 3 维图草图，以促进图流中的快速准确查询。DMatrix 支持结构查询和基于权重的估计。通过代表密钥保留和多数投票的结合，DMatrix 可以有效地缩小查询的错误范围，并具有实时响应效率。理论分析和实验结果都证实，与最先进的技术相比，我们的解决方案在准确性和效率上都更胜一筹。