3.3. Updating WEPS-Tree
In this section, we describe the contents of tree updates corresponding to Phase I. Here, we introduce and define our own tree and list data structures used in the proposed algorithm, and devise tree update methods including tree constructing and restructuring techniques. In particular, empirical examples are followed by the techniques.
Frequent pattern mining approaches based on the sliding window model [27] and [37] use their own tree structures to store real-time data streams and mine patterns. Similarly, our algorithm employs such a basic framework in order to process continuous product data streams. In addition, the proposed algorithm utilizes not only a tree structure but also a list data structure including essential information extracted from the tree. The details of these data structures are defined and explained in this section. They also allow the proposed algorithm to reduce the number of database scans by a single time and mine weighted erasable patterns more efficiently. The tree structure, called WEPS-Tree, is composed as follows. In this algorithm, WEPS-Tree is used for efficiently storing the latest information of a given product data stream and preparing the mining process. WEPS-Tree is composed of a header table and a prefix tree. The header table is employed to store the abridged information of the tree and decide its order. Each node of WEPS-Tree includes the six fields: an item name, a profit, a pre-order index, a reconstruction flag, a set of child node links, and a parent node link. The item name and the profit are the corresponding node’s name and a sum of all the profits related to the node, respectively. The pre-order index is a number inputted through the tree traversal in a pre-order manner, and the reconstruction flag is a variable for checking whether or not the corresponding node has been restructured. The parent and child node links are pointers to connect the node to its parent and children, respectively. In addition, each node is divided into the two types of nodes, an ordinary node and a tail node. The tail node also includes an additional variable that stores pane information for controlling the sliding window model. This variable has a list form, and its size is decided by the number of panes contained into a window. Once the window and pane sizes are decided, PID and pane information is automatically managed. If the window and pane sizes are 3 and 2, respectively, three elements are created in each tail node. After that, transactions are inserted into the tree in sequence; at the same time, the profit value of each transaction is stored into the corresponding element of the tail node. Pane information is accumulated until the sliding window is filled with transaction data. Thereafter, data belonging to the oldest pane (i.e., the first element of each tail node) are deleted. In this process, the corresponding pane information is also removed, and the remaining pane information is shifted to the left.
3.3. การ WEPS ต้นไม้ในส่วนนี้ เราอธิบายเนื้อหาของการปรับปรุงแผนภูมิที่สอดคล้องกับขั้นตอนที่ผม ที่นี่ เราแนะนำ และกำหนดแผนภูมิของเราเอง และโครงสร้างรายการข้อมูลที่ใช้ในอัลกอริทึมที่เสนอ และต้นไม้ประดิษฐ์ปรับปรุงวิธีการสร้างต้นไม้และโครงสร้างเทคนิค โดยเฉพาะ ตัวอย่างเชิงประจักษ์จะตาม ด้วยเทคนิคการFrequent pattern mining approaches based on the sliding window model [27] and [37] use their own tree structures to store real-time data streams and mine patterns. Similarly, our algorithm employs such a basic framework in order to process continuous product data streams. In addition, the proposed algorithm utilizes not only a tree structure but also a list data structure including essential information extracted from the tree. The details of these data structures are defined and explained in this section. They also allow the proposed algorithm to reduce the number of database scans by a single time and mine weighted erasable patterns more efficiently. The tree structure, called WEPS-Tree, is composed as follows. In this algorithm, WEPS-Tree is used for efficiently storing the latest information of a given product data stream and preparing the mining process. WEPS-Tree is composed of a header table and a prefix tree. The header table is employed to store the abridged information of the tree and decide its order. Each node of WEPS-Tree includes the six fields: an item name, a profit, a pre-order index, a reconstruction flag, a set of child node links, and a parent node link. The item name and the profit are the corresponding node’s name and a sum of all the profits related to the node, respectively. The pre-order index is a number inputted through the tree traversal in a pre-order manner, and the reconstruction flag is a variable for checking whether or not the corresponding node has been restructured. The parent and child node links are pointers to connect the node to its parent and children, respectively. In addition, each node is divided into the two types of nodes, an ordinary node and a tail node. The tail node also includes an additional variable that stores pane information for controlling the sliding window model. This variable has a list form, and its size is decided by the number of panes contained into a window. Once the window and pane sizes are decided, PID and pane information is automatically managed. If the window and pane sizes are 3 and 2, respectively, three elements are created in each tail node. After that, transactions are inserted into the tree in sequence; at the same time, the profit value of each transaction is stored into the corresponding element of the tail node. Pane information is accumulated until the sliding window is filled with transaction data. Thereafter, data belonging to the oldest pane (i.e., the first element of each tail node) are deleted. In this process, the corresponding pane information is also removed, and the remaining pane information is shifted to the left.
การแปล กรุณารอสักครู่..
