사용자 클러스터링을 위한 개량된 K-평균 알고리즘의 구현 = An Improved K-Means Algrithm for User Clustering

다국어 초록 (Multilingual Abstract)

The user clustering for web navigation pattern discovery is very useful to get preference and behavior pattern of users for web page, And also, such informations are very essential for web personalization or customer grouping.
In this thesis, an algorithm for clustering the user's web navigation path is proposed and then some special navigation pattern can be recognized by the results of algorithm. The proposed algorithm has two clustering phases.
In the first phase, all paths are classified into k-groups on the bases of the their similarities. In order to establish the similarity measure, each path is transformed into a feature vector, and as angle between two feature vectors is used as a measure of the similarity. In the measure, the smaller angle a pair of vectors have, the more similarity is guaranteed. The solution obtained in the first phase is not global optimum but it gives a good and feasible initial solution for the second phase.
In the second phase, the first phase solution is improved by the revised k-means algorithm, in which grouping the paths is performed by the hyperplane instead of the distance between the each path and the group center. Computational results show that the proposed method is more efficient and time-saving, but it requires huge computer memory spaces.