In many applications, databases are frequently changed by insertions, deletions, and/or modifications of transactions. Consequently, the frequent patterns extracted from them must be updated. Researchers propose incremental mining to update frequent patterns efficiently instead of mining all frequent patterns from scratch. Although FP-tree is one of the most efficient algorithms for frequent pattern mining, it is not easily adoptable with incremental updating. Accordingly, the CP-tree and restructuring method of Branch-Sorting have been proposed for incremental mining of frequent pattern. This method consists of two main phases of insertion and restructuring. Since during construction of CP-tree items are sorted in descending order of previous insertion phase, then its restructuring can be very costly. To solve this weakness, in this paper a new efficient prefix tree structure has been proposed to reduce the time of restructuring. The proposed tree is created based on the frequency of last items and it requires just one database scan. The experimental results show that using the proposed tree and Branch-Sorting method can enhance the efficiency of incremental mining of frequent patterns from both dense and sparse datasets.
展开▼
