BIGDATA: Collaborative Research: F: Association Analysis of Big Graphs: Models, Algorithms and Applications

点击翻译按钮获取中文摘要

Association analysis is a fundamental problem in Big Data analytics. Emerging applications require computationally efficient association models and scalable association mining techniques to find regularities of graph data. Conventional association analysis for transactional data is hard or infeasible to be adapted to effectively support the next generation of graph data analytics, especially under limited computing resources. In this project, the PIs develop models, algorithms and tools to support association analysis over large-scale graph data under resource constraints. The project formulates new variants of the conventional association model that are enhanced by advanced capability of graph queries. Both exact and approximate querying and mining paradigms are explored to support effective association analysis over multi-source, large-scale, and fast-changing graph data. The PIs instantiate the generic framework to two practical association analysis scenarios, notably, a) multi-graph association analysis, and b) association detection over graph streams. The project develops a package of distributed and stream association mining techniques supported by the proposed generic model and algorithms.The enhanced model and algorithms enable scalable association analysis in a wide range of massive data applications. The principles learned from this project can be applied to big data analytics and system design in general. The study of new association analysis framework has immediate applications in emerging areas, including data quality, affinity marketing, and network security. Application collaborators of the project include Pacific Northwest National Laboratory, LogicMonitor, and Facebook. Broader impacts of the project also include research training and education of students including women and minorities, and design of new curricula and education tools that target both CS and non-CS students.