报告时间：2015年7月24日星期五下午3:00pm
报告地点：浙江大学紫金港校区蒙民伟楼CAD&CG国家重点实验室402室
报告题目：Mining Spatio-Temporal Correlation Patterns from Data Sources across Different Domains
报告人：Dr. Jie Bao
主持人：巫英才 研究员
Abstract: In the big data era, we are facing a huge amount of spatio-temporal data generated from different domains, such as transportation, urban environments and social media. Moreover, these heterogeneous data sources may interfere with each other and result in correlation patterns, concerned with a certain spatio-temporal constraint. For example, when the following three conditions occur within 3 km and last for two hours, a pattern with three different data sources may be discovered as: traffic speed < {30km/h}, weather = {foggy}, air quality = {unhealthy}. Such cross domain spatio-temporal correlation suggests the underlying connection between different domains, providing rich contexts and indications for a diversity of applications, e.g., air pollution diagnoses, business location selections, and context-aware recommendations.
Due to the scale, diversity and heterogeneity of cross domain data sources, however, extracting such patterns is far beyond human capability and the scope of existing association rules mining. To this end, we propose a generic framework to automatically extract the frequent spatio-temporal correlation patterns from data sources across different domains, which overcomes the challenges of: various value scales (numeric and categorical), different data update frequencies, and exponential size of possible pattern combinations. Our framework employs three main modules: 1) spatio-temporal indexing, which creates an index for cross domain data sources; 2) pre-processing, which partitions the data and prunes the impossible data source combinations; and 3) pattern mining, which uses the value matrix to hold different scale combinations of values, and applies a sweep-line based algorithm to efficiently identify the patterns. We test our system with the real data sources from Beijing. The results confirm that our system discovers interesting patterns, and significantly outperforms the baseline approach.
Short Bio: Dr. Jie Bao, currently, is an associate researcher at urban computing group in MSR Asia. His research interests lie broadly over spatio-temporal data management and location-based services. Before joining MSRA, he got his BS, MS, and Ph.D. degree in Computer Science from Zhejiang University, Auburn University and University of Minnesota at Twin Cities, respectively.
Menu
