A lot of real world problems can be modeled as traversals on graph. Mining from such traversals has been found useful in several applications. However, previous works considered only unweighted traversals. This paper generalizes this to the case where traversals are given weights to reflect their importance. A new algorithm is proposed to discover frequent patterns from the weighted traversals. The algorithm adopts the notion of confidence interval to distinguish between confident traversals and outliers. By excluding the outliers, more reliable frequent patterns can be obtained. In addition, they are further ranked according to their priority. The algorithm can be applied to various applications, such as Web mining.
展开▼
