Mining Matrix Pattern from Mobile Users - Distributed Artificial Intelligence, Agent Technology, and Collaborative Applications - page 254

r ela ted work
Recent related work done by others in the field of data mining includes the dense region finding (Yip,
Wu, Ng, & Chan, 2004), finding negative rules (Thiruvady & Webb, 2004), secure association rules
sharing (Oliveira, Zaiane, & Saygin, 2004), discovery of maximally frequent pattern tree (Miyahara,
Suzuki, Shoudai, Uchida, Takahashi, & Uedal, 2004), spatial association rules (Koperski & Han, 1995)
and time association rules (Barbar'a et al., 2004). It can be observed that data mining (Fayyad, 2004) is
widely recognised and research in the field has tried to extend data mining into different areas, such as
time series (Han et al, 1999; Han et al., 1998) and geographical (Goh & Taniar, 2004; Koperski & Han,
1995). In our case, we have focused on the area of mobile user data mining, which involves gathering
source data from mobile users and perform data mining (Goh & Taniar, 2004a, 2004b, 2005). Mobile
user data mining is still at an infant stage, with only a few works having been done. One significant
work done by others is group pattern (Wang et al., 2003) which aims to find the group characteristics
of mobile users. Group pattern contains some inefficiency, including high amount of processing power
required, and weaknesses, such as only physical domains of the problem is being observed. Our previ-
ous work in frequency patterns (Goh & Tanair, 2004a) aims to solve this shortcoming.
Frequency pattern is one of our existing methods. Frequency pattern (Goh & Tanair, 2004a) is a
method for finding relationships between the mobile users by means of observing the frequency of
communication in between these mobile users. This method was developed to enhance group pattern
(Lim, Wang, Ong, et al., 2003), that examines the group relationships of mobile users by using physical
distance which does not address the fundamental challenge of mobile environment, that is to stay in
touch without distance barrier. Frequency pattern (Goh & Tanair, 2004a) solves this problem by us-
ing frequency of communication instead of physical distance to calculate closeness of each individual
mobile user.
Parallel pattern (Goh & Taniar, 2005) is another one of our existing methods. Parallel pattern aims
to find out similarities of decision among mobile users. The similarities of decision can be separated
into physical decision and logical decision (Goh & Taniar, 2005). Parallel pattern enables the ability to
find out the trends of the movement pattern of mobile users (Goh & Taniar, 2005), by examining either
how they move from one location to another, or through examining how they move from one context
to another.
f frequency Pattern
Frequency pattern (Goh & Taniar, 2004a) is designed to use frequency of communication between each
mobile user, coupled with the pre-specified criteria (Goh & Taniar, 2004a) in order to further enhance the
accuracy of frequency pattern. The pre-specified criteria (Goh & Taniar, 2004a) configure the method
in order to allow the decision maker to place different amount of emphasis on different parts of time
zone concerned. For instance, in the business environment, mobile communications that occur most
recently serve as the strongest indication of relationship. This is realised by having the pre-specified
criteria to configure high emphasis on the most recent communications.
Frequency pattern (Goh & Taniar, 2004a), therefore, uses pre-specified criteria in order for the
decision maker to place different emphasis on different time zones of the window size. Sometimes, it
not the recent communications that need to be taken into consideration but somewhere just before the
recent communication. The ability to adjust and place different emphasis at different parts of the time