Research On Association Rules Mining Algorithm Based On Positive & Negative Items And Multilateral Support

As an important sub-branch of data mining, association rules have been successfully applied in many fields now. However, most of the association rules consider only the positive contact between transaction attributes. The negative contact hidden in the data did not attract much attention.This thesis describes an integrate form of association rules. Taking into account the positive items and negative items in the transaction database, this form of association rules not only reflects the positive contact between transaction attributes, but also reflects the negative contact between transaction attributes.Firstly, to address the issue that the number of frequent itemsets will greatly increase because of considering the negative items, this thesis proposes an improved algorithm, named as MFP_PN, based on a detailed study of the FP-growth algorithm. The novel algorithm inherites the advantage of the FP-growth algorithm that does not need to scan the database repeatedly, and takes into account the positive items and negative items without expanding the original database. In addition, the new algorithm builds Tree_PN with positive and negative items by following the FP-tree structure. Mining frequent itemsets through extending frequent patterns based on conditional frequent suffixal items, the novel algorithm does not need to construct a large number of conditional model trees, and saves time and space. The tests show that MFP_PN has better performance than FP-growth algorithm.Secondly, the thesis analyzes the limitations of minimum support and the usefulness of the items with oversize support, and proposes a model with positive and negative multilateral support, i.e., PNMS model. The model can not only rule out the meaningless items with oversize support, but also adjust the number of positive and negative items in the mining results through setting different positive and negative minimum support threshold in order to meet the different needs of customers. The experiments show that PNMS model is effective and feasible in improving the usefulness of the rules and adjusting the number of positive and negative items.