Research And Improvement Of Algorithm For Incremental Updating Association Rule In Retail Business Intelligence System

Data Mining, as one of the key techniques in database, is an important method which discovers the valuable information from massive sales transactions data. Association rules mining is a primary research interest, especially for the decision support in the retail industry, which has precious research value and wide application prospects.This dissertation mainly focus on study and improvement of the algorithm for updating association rules with incremental transactions and minimum support changes simultaneously, which is based on MVS Retail Business Intelligence System(MRBIS).Compared with positive association rules mining, the study of negative association rules is less but it is also important for the decision support in the retail industry. Therefore, this dissertation also focus on the algorithm for updating negative association rules based on the study of negative association rules concepts. This dissertation mainly consists of three parts:(1) Incremental updating association rules is only for incremental transactions or minimum support changes. This dissertation proposes a new algorithm FIM_AIUA, which updates association rules with incremental transactions and minimum support changes simultaneously. The algorithm expands FIM algorithm and AIUA algorithm, improves the efficiency and corrects the mistakes of My_IUA algorithm. Moreover, it modifies FIM algorithm with a new argument and presents a new function fim_aiua_gen0that rewrites the function aiua_gen() of AIUA algorithm. Experiments with real transaction data of MRBIS show that FIM AIUA is efficient and outperforms both My_IUA algorithm and Apriori algorithm.(2) Incremental updating association rules requires the frequent itemsets of the updated database.Negative association rule updating requires not only frequent itemsets but also infrequent itemsets.Therefore the algorithm needs to find all frequent and infrequent itemsets. This dissertation proposes two algorithms NAIUA and NIUA_NAIUA. NAIUA expands AIUA to updating negative association rules with minimum support changes and NIUA_NAIUA expands NIUA and NAIUA to updating negative association rules with incremental transactions and minimum support changes simultaneously.(3) This dissertation is based on MRBIS and all three algorithms proposed in this dissertation are implemented in this system to update association rules with incremental transactions and minimum support changes. The experiments that use POS data from MRBIS show the efficiency of these algorithms, which improve the efficiency of MRBIS association rules updating.