| Summary: | 碩士 === 銘傳大學 === 資訊工程學系碩士班 === 96 === Traditional association rule mining is to find association rules in a given transaction database. However in many real applications, there are a lot of new data generated continuously and the user wants to get the new association rules in the updated transaction database. Hence how to find association rules efficiently under the circumstances where data would be added continuously becomes an important research issue for a practical purpose. It is called incremental mining of association rules or mining association rules in data streams.
Many incremental mining algorithms and data streams mining algorithms proposed recently adopt the tree-based structure. It constructs a tree structure to store transactions in memory and then uses an algorithm similar to FP growth to find frequent itemsets from the tree structure. When the incremental transactions arrive and we want to find frequent itemsets in the updated transaction database, it would add the incremental transactions to the tree structure constructed previously and then uses the algorithm similar to FP growth to mine frequent itemsets from the updated tree structure again. However the size of the original transaction database is often much larger than that of incremental transactions. So it would be inefficient. In the paper, we propose an efficient algorithm for incremental mining of frequent patterns. When mining from the updated transaction database, it mines frequent itemsets from the incremental transactions and then integrates the mining result with the previous result to get the new frequent itemsets in the updated transaction database. Finally, we can use these new frequent itemsets to generate new association rules. Experiment results show our approach is more efficient.
|
|---|
