利用9DLT字串表示法找尋空間關聯規則方法

Abstract

在網路多媒體資料逐步成長的現在，空間資料挖掘(Spatial data mining)，扮演著重要角色，它能發掘空間資料庫中的隱性知識、物件間空間關聯亦或找尋出令人感興趣的樣式。這篇論文著重於找尋物件間空間關聯規則，過去已有學者提出使用viewpoint mining 方法與使用2D 字串表示法找尋co-location rules 的方法。然而，實驗結果可能過於詳細亦或物件間空關關係的描述過於模湖不清。 因此，我們提出一個新的演算法「9DLT-Miner」去找尋空間關聯規則，並利用9DLT 資料表示方式描述每一個影像。9DLT-Miner 採用Apriori 方法的概念、anti-monotone 與9DLT 的過濾策略。9DLT-Miner 分為兩個階段。第一階段，找出所有長度1 的frequent 樣式。第二階段，使用長度k (k>=1)的樣式去產生所有長度k+1 的候選樣式並計算其support 數，以確認是否為frequent 樣式。反覆上 述步驟，直到找不出任何frequent 樣式為止。從實驗結果顯示，9DLT-Miner 能有效過濾大量不可能的候選樣式，可節省大量時間。

Nowadays, there are the increasing numbers of images accumulated on the Internet. Spatial data mining play an important role of extracting implicit knowledge, spatial relationships among objects and other interesting patterns stored in spatial databases. In this thesis, we focus on finding the association rules of spatial relations among objects in an image. Previously, some scholars have proposed viewpoint mining and co-relation mining method based on the 2D string representation. However, the mining results may be too detailed and the relations between the objects are vague. Therefore, we propose a novel algorithm, 9DLT-Miner, where every image is represented by the 9DLT representation. 9DLT-Miner adopts the concept of the Apriori algorithm as well as uses the anti-monotone and 9DLT pruning strategies. Our proposed method consists of two phase. In the first phase, we find all frequent patterns of length one. In the second phase, we use the frequent k-patterns (k>=1) to generate all candidate (k+1)-patterns and then scan the databases to count the support and check if a pattern is frequent. Repeat the steps in phase 2 until no more frequent patterns can be found. Experimental results show that 9DLT-Miner prunes a large number of impossible frequent candidates, and it’s more efficient and scalable.