Mining Subspace Skylines with a Sliding Window over Data Streams

• Main Authors: Yi-Shien Liu, 劉益先
• Other Authors: Yi-Hung Wu
• Format: Others
• Language: zh-TW
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/02724529313297100794

碩士 === 中原大學 === 資訊工程研究所 === 98 === In multi-dimensional data applications, skyline objects, classified as full skyline and subspace skylines, are especially outstanding compared with the other objects. The full skyline takes all the dimensions into account, while the subspace skylines consider only part of the dimensions. We research into the problem of mining subspace skylines with a sliding window over data streams. As the objects move in the multi-dimensional space, the skyline objects vary as time goes. Thus, we need to compute the differences of dimension values among objects to keep track of the skyline objects in every subspace. In real cases, non-skyline objects are in the majority. To avoid unnecessary computations on some non-skyline objects, we record the full skyline objects that dominate them. Besides, according to the dominance and coincidence relationships among the full skyline objects, we employ logical operations to compute the subspaces in which they are also skyline objects. Furthermore, the non-skyline objects that are subspace skylines can also be discovered from the recorded information of the full skyline objects dominating them. The experimental results show that our method for avoiding unnecessary computations on non-skyline objects can reduce on average 30% of the execution time and achieve the accuracy above 90%. The mining of subspace skylines performs well especially when the number of dimensions is low.