MUSK: Uniform sampling of k maximal patterns

摘要: Recent research in frequent pattern mining (FPM) has shifted from obtaining the complete set of patterns to generating only a representative (summary) subset patterns. Most existing approaches this problem adopt two-step solution; first step, they obtain all patterns, and second some form clustering is used summary set. However, twostep method inefficient sometimes infeasible since step itself may fail finish reasonable amount time. In paper, we propose an alternative approach representatives based on uniform sampling output space. Our new algorithm, Musk, obtains by uniformly pool maximal patterns; uniformity achieved variant Markov Chain Monte Carlo (MCMC) algorithm. Musk simulates random walk partial order graph with prescribed transition probability matrix, whose values are computed locally during simulation. stationary distribution walk, nodes sampled uniformly. Experiments various kind itemset databases validate effectiveness our approach.