Study On Algorithm Of Mining Association Rules With Negative Items

In recent years, with development of computer technology, data management technology was successfully applied, and enterprise information continuously improves. Vast amounts of data accumulate in the database on various application fields. Using data mining techniques can access to the correct, interesting and potentially valuable knowledge from the large-scale data. Mining Association Rules is one of the significant research methods, and have important theoretical value and broad application prospects. At present, the research field mainly focuses on the traditional positive association rules.The research to the negative association rules has not been attracted. However, in many fields, only mining positive association rules is not enough, and it's need to mining the denial of the data. It is therefore necessary to study the relations between negative attributes.There are only a few algorithm of mining negative association rules, and most of them are based on Apriori idea which need to scan data sets for many times, and generated a large number of candidate frequent itemsets. Due to the limitations, the paper presents a frequent itemsets mining algorithm which is based on the frequent pattern tree. Algorithm uses the data structure of FP-growth algorithm to compress and store transactions. So it has no need to repeatedly scanning the database, and not the generation of large amount of candidates'itemsets. In addition, we add a bit string to each node in the tree to store the prefix of the items. By using the structure we can avoid repeatedly traversing the sub-tree while expanding the patterns. The experiments show that, the algorithm the paper presents has better efficiency than similar existing mining algorithm.In addition, while mining association rules with negative items, because of the introduction of negative item, the searching space of frequent itemsets has become greater, and at the same time the number of negative association rules also increased. However, many of the rules are not interested for the users, and may even contain errors. Therefore, on the basis of traditional "support - Confidence - Interest" evaluation criteria, the maximum support is introduced. While mining frequent itemsets, the meaningless itemsets will be eliminated. The experiments show that the improved algorithm is correct and effective.