Mining association rules between sets of items in large databases
SIGMOD '93 Proceedings of the 1993 ACM SIGMOD international conference on Management of data
Efficiently mining long patterns from databases
SIGMOD '98 Proceedings of the 1998 ACM SIGMOD international conference on Management of data
Mining frequent patterns without candidate generation
SIGMOD '00 Proceedings of the 2000 ACM SIGMOD international conference on Management of data
Scalable Algorithms for Association Mining
IEEE Transactions on Knowledge and Data Engineering
Discovering All Most Specific Sentences by Randomized Algorithms
ICDT '97 Proceedings of the 6th International Conference on Database Theory
Mining All Non-derivable Frequent Itemsets
PKDD '02 Proceedings of the 6th European Conference on Principles of Data Mining and Knowledge Discovery
Fast Algorithms for Mining Association Rules in Large Databases
VLDB '94 Proceedings of the 20th International Conference on Very Large Data Bases
Computational complexity of itemset frequency satisfiability
PODS '04 Proceedings of the twenty-third ACM SIGMOD-SIGACT-SIGART symposium on Principles of database systems
Proceedings of the twenty-fourth ACM SIGMOD-SIGACT-SIGART symposium on Principles of database systems
An efficient rigorous approach for identifying statistically significant frequent itemsets
Proceedings of the twenty-eighth ACM SIGMOD-SIGACT-SIGART symposium on Principles of database systems
An Efficient Rigorous Approach for Identifying Statistically Significant Frequent Itemsets
Journal of the ACM (JACM)
Hi-index | 0.00 |
We study the relative effectiveness and the efficiency of computing support-bounding rules that can be used to prune the search space in algorithms to solve the frequent item-sets mining problem (FIM). We develop a formalism wherein these rules can be stated and analyzed using the concept of differentials and density functions of the support function. We derive a general bounding theorem, which provides lower and upper bounds on the supports of item-sets in terms of the supports of their subsets. Since, in general, many lower and upper bounds exists for the support of an item-set, we show how to the best bounds. The result of this optimization shows that the best bounds are among those that involve the supports of all the strict subsets of an item-set of a particular size q. These bounds are determined on the basis of so called q-rules. In this way, we derive the bounding theorem established by Calders [5]. For these types of bounds, we consider how they compare relative to each other, and in so doing determine the best bounds. Since determining these bounds is combinatorially expensive, we study heuristics that efficiently produce bounds that are usually the best. These heuristics always produce the best bounds on the support of item-sets for basket databases that satisfies independence properties. In particular, we show that for an item-set I determining which bounds to compute that lead to the best lower and upper bounds on freq(I) can be done in time O(|I|). Even though, in practice, basket databases do not have these independence properties, we argue that our analysis carries over to a much larger set of basket databases where local "near" independence hold. Finally, we conduct an experimental study using real baskets databases, where we compute upper bounds in the context of generalizing the Apriori algorithm. Both the analysis and the study confirm that the q-rule (q odd and larger than 1) will almost always do better than the 1-rule (Apriori rule) on large dense baskets databases. Our experiment reveal that on these baskets databases, the 3-rule prunes almost 100% of the search space while, the 1-rule prunes 96% of the search space in the early stages of the algorithm. We also observe a reduction in wasted effort when applying the 3-rule to sparse baskets databases. In addition, we give experimental evidence that the combined use of the lower and upper bounds determine the exact support of many frequent item-sets without counting.