Article Preview
Top1. Introduction
Mobile web services pattern mining is an important research topic nowadays. These services are light weighted applications, which are used for performing a specific task. These web services are accessed using the internet via smart phones or laptops. A particular user may access, a series of services at different times at different locations or a single location. To extract the interesting pattern of services, data mining techniques are used. By sequential pattern mining (Agrawal & Srikant, 1995; Mohbey & Thakur, 2015) web services sequence can be extracted. These sequences are helpful to find the behavior of a specific user. The generated mobile web service patterns are used in different fields like behavior analysis of users, finding most accessed services occurrences, publishing a new service, promoting business, etc. Figure 1 shows the simple scenario in which different mobile web services are accessed by the mobile users at different locations. In this figure L1, L2, L3... Lm defined for different locations and W1, W2, W3... Wn for mobile web services. Here S1, S2, S 3... Sp represent different services accessed sequences. The traditional sequence pattern mining approach only considers the items (Lan et al., 2014). It does not include any constraint or factor like price, profit or preferences of items. Sometime the low frequency of items or service may be important. For example, assume there exists a pattern <mail, news> in a sequence and assume it is a low frequency pattern in the sequence database. However, this pattern may contribute a large portion to the overall profit of the service provider (Hwang et al., 2013). To handle this, Yun et al. proposed a new research issue, namely weighted sequential pattern mining (Yun & Leggett, 2006) in which different weights were assigned to items by the importance of each item.
Figure 1. Mobile web services sequences
To evaluate the weight value of a sequence pattern, Yun et al. designed an average weight function (Yun & Leggett, 2006). Yun et al. (Yun & Leggett, 2006) also developed an upper bound model, in which maximum weight of all sequences is used as the upper bound weight value. This value is used to construct a downward closure property in the problem of weighted sequential pattern mining. Yun et al. proposed WSpan algorithm (Yun & Leggett, 2006) to avoid information loss in mining. A large number of the candidate sub sequence was still generated due to the upper bound of overestimated weight values for the candidate. Lan et al. (Lan et al., 2014) proposed an approach for finding weighted sequential patterns. They apply this approach on the traditional transaction and items. To address the above reason, we proposed a utility based approach to reduce the large number of candidate generation. Here we have used a utility value for each mobile web service based on the accessed preferences. The major contributions of this proposed work are summarized as follows. In the proposed work we have used an efficient sequence maximum utility (SMU) approach for strong upper bound of utility support in sub sequences. We have proposed UMWSPM (Utility based Mobile Web Service Pattern Mining) approach for finding interesting mobile web services patterns. The proposed approach speeds up the execution efficiency in finding utility based patterns.