Efficient Approaches for Updating Sequential Patterns

Abstract

Mining sequential patterns is to find the sequential purchasing behaviors for most of the customers in a transaction database. By using sequential patterns, it is possible to predict which products will be purchased in the future after the customer purchases certain commodities. Nowadays, transaction data is continuously added to the database. It is an important issue to update the sequence pattern efficiently in this environment. The previous efficient approach is to store the transactions in a tree structure. When the transactions were added, the tree structure could be updated according to the newly added items. It still needs to re-find the sequential patterns from the updated tree structure and re-scan the original transactions, without using the previous patterns. Therefore, we propose two algorithms for mining and maintaining the discovered sequential patterns when the transactions are added into the database. Our algorithms do not need to re-scan the original transactions and re-generate the existing sequential patterns, which just need to process the added transactions to update the existing sequential patterns. The experimental results also show that our algorithms outperforms the previous approaches.