改进的频繁和高效用项集挖掘算法

Improved Mining Algorithm for Frequent and High Utility Itemsets

提出一种基于局部效用质量值的上界剪枝新方法,引入伪投影技术避免真实地构造物理投影,基于二者提出改进的FHIMA-P算法.在提出的FHIMA-P算法中引入事务合并和投影事务合并技术,提出最终的FHIMA-MP算法,并在mushroom和accident数据集上进行实验.结果表明:FHIMA-P算法的运行时间相比FHIMA-ALL算法缩短,而FHIMA-MP算法则较前两者效率有非常大的提高;在不同参数下,mushroom和accident数据集中大量可合并事务(投影事务)数目也很好地证明了事务(投影事务)合并的有效性.

A new method that uses the upper bound of quality to prune the search space based on local utility quality is proposed,meanwhile,pseudo projection technique is introduced to avoid actually construct the physical projection,then based on these two points,an improved FHIMAP algorithm is proposed.By adding the transaction merging and projected transaction merging technique in FHIMAP algorithm,the final FHIMAMP algorithm is proposed.An experiment is conducted on mushroom and accident dataset,the result shows that the running time of FHIMAP algorithm is shorter than that of FHIMAALL algorithm,while the FHIMAMP algorithm improves significantly compared with the previous two algorithms′efficiency.Moreover,the huge number of transactions(projected transaction)that can be merged on mushroom and accident dataset in different papameters also prove the effectiveness of transaction(projected transaction)merging technique.