|
Title:
|
MINING FIRST-COME-FIRST-SERVED FREQUENT TIME SEQUENCE PATTERNS IN STREAMING DATA |
|
Author(s):
|
Atsushi Okamoto, Takayoshi Shoudai |
|
ISBN:
|
978-972-8939-82-3 |
|
Editors:
|
Piet Kommers and Pedro Isaías |
|
Year:
|
2013 |
|
Edition:
|
Single |
|
Keywords:
|
Data Mining, Streaming Algorithm, Frequent Pattern Mining, Time Sequence Pattern, Lossy Counting. |
|
Type:
|
Full Paper |
|
First Page:
|
283 |
|
Last Page:
|
290 |
|
Language:
|
English |
|
Cover:
|
|
|
Full Contents:
|
click to dowload 
|
|
Paper Abstract:
|
In this paper, we discuss the data mining problem of finding frequent time sequence patterns of constant length in a stream data. A time sequence pattern is an alternating finite sequence of events and positive integers (e1,T1,e2,T2,
,Tk-1,ek). It represents that the i-th event ei is followed by the (i + 1)-th event ei+1 within Ti events for i = 1,
,k-1. To count the frequency of a time sequence pattern effectively, we define the first-come-first-served (FCFS)-maximal frequency, which is a natural frequency according to the FCFS rule for a stream data. We propose a round-robin and an Apriori-like lossy counting algorithm for finding all frequent time sequence patterns with respect to FCFS-maximal frequency, and show that the round-robin algorithm always maintains an ?-deficient synopsis and that the Apriori-like algorithm maintains it for k ? 3. Finally, we present experimental evaluations of our algorithms on real data. |
|
|
|
|
|
|
