In order to mine weighed erasable patterns over sliding
window-based data streams, the proposed algorithm performs
unique mining processes as follows. Fig. 2 shows the overall
architecture of the proposed algorithm. All the processes are
mainly divided into two phases. In Phase I, the algorithm performs
tree constructing and restructuring tasks for the current window,
and in Phase II, its mining operations are conducted to discover
weighted erasable patterns from candidate patterns. WEPS creates
a window and stores the latest data in a given product data stream,
where the proposed tree structure, WEPS-Tree, is constructed on
the basis of the window data (see Section 3.3.1). The constructed
WEPS-Tree continues to be updated according to the accumulation
of the product data stream (see Section 3.3.2). After the tree
construction, each path of the tree is sorted in a support descending
order through a tree restructuring step. Thereafter, the algorithm
generates Node-List for the weighted erasable pattern mining,
traversing the restructured WEPS-Tree (see Section 3.3.1). Using the
information of Node-List, candidates of weighted erasable patterns
with length 1 are generated and then pattern expanding operations
are conducted from them (see Section 3.4.1). In this process, the
algorithm performs pattern combinations in a recursive manner
and extracts weighted erasable patterns from the candidates by
using various factors.