Surprising results of trie-based FIM algorithms

IEEE ICDM Workshop on Frequent Itemset Mining Implementations

Ferenc Bodon

Others  :  http://CEUR-WS.org/Vol-126/bodon.pdf
PID  :  1785

来源: CEUR

【 摘 要 】

Trie is a popular data structure in frequent itemset mining (FIM) algorithms. It is memory-efficient, and allows fast construction and information retrieval. Many trie-related techniques can be applied in FIM algorithms to improve efficiency. In this paper we propose new techniques for fast management, but more importantly we scrutinize the well-known ones especially those which can be employed in APRIORI. The theoretical claims are supported by results of a comprehensive set of experiments, based on hundreds of tests that were performed on numerous databases, with different support thresholds. We offer some surprising conclusions, which at some point contradict published claims.