标题: | 大型时空资料库中拓朴样式探勘之渐进式维护 Incremental Maintenance of Topological Patterns in Large Spatial-Temporal Database |
作者: | 吴昭莹 Wu, Chao-Ying 李素瑛 Lee, Suh-Yin 多媒体工程研究所 |
关键字: | 资料探勘;增量式探勘;拓扑样式;时空资料库;时间样式;data mining;incremental mining;topological pattern;spatial-temporal database;collocation pattern |
公开日期: | 2011 |
摘要: | 在许多空间时间资料库的生活应用,例如环境生态分析、气象分析、位置基础分析,大都随着时间变化做增量的更新。当资料库增量更新后,有些已发现的拓扑样式会无效,而有些新的拓扑样式会出现。 当新的事件加入资料库,假如每一次的更新都必须重新探勘拓扑样式,将是一件既没效率且不切实际的工作。尽管最近有学者提出维护拓扑样式的方法,而且我们也可以应用既有探勘静态资料库的演算法重新探勘更新后的资料库。然而,既存的演算法并不是非常有效率。 在大型时空资料库中拓扑样式探勘之渐进式维护是一件艰钜的工作,因为拓扑样式探勘相较一般项目集样式是比较复杂的。在这篇论文,我们提出一个演算法,Inc_TMiner,主要是设计在增量的时空资料库中维护拓扑样式。在合成资料的实验结果显示 Inc_TMiner 在执行时间优于之前的渐进式演算法,也优于利用现有探勘静态资料库的演算法重新探勘更新后的资料库。 Spatial temporal data mining is an important research area with many interesting topics, such as ecology analysis, meteorology analysis, location-based analysis and so forth. Most spatial temporal databases are updating incrementally with time. Some discovered topological patterns may be invalidated and some new topological patterns may be introduced by the evolution of databases. When new instances are inserted into the database, we can re-mine topological patterns from scratch each time using the existing static algorithms. Some researches on the maintenance of topological patterns in an incremental manner are proposed. However, all static algorithms and incremental algorithms are incompetent and not scalable. In this thesis, an efficient algorithm, Inc_TMiner (Incremental Topology Miner) is developed to incrementally maintain topological patterns from spatial-temporal databases. The experimental results on synthetic datasets indicate that Inc_TMiner significantly outperforms the static algorithms and the existing incremental algorithm in execution time and possesses graceful scalability. |
URI: | http://140.113.39.130/cdrfb3/record/nctu/#GT079757515 http://hdl.handle.net/11536/46055 |
显示于类别: | Thesis |
文件中的档案:
If it is a zip file, please download the file and unzip it, then open index.html in a browser to view the full text content.