The Research And Application Of Weighted Association Rules Mining Algorithm

Data mining can acquires knowledge and rules that are implicit, unknown and having potential value for decision-making from large databases or data warehouses. Association rule mining is a very important research field in data mining,with the the purpose of finding the relations among items in the databbase. From the standpoint of whether to generate frequent itemsets,association rule mining algorithm can be divided into the algorithm of generating frequent items and not generating frequent items.The classical algorithms are Apriori Algorithm and FP-growth Algotithm,but neither considers the different importance of the items in the database.This thesis investigates the algorithm of weighted association rules.The main work and innovations of this thesis are are as follows:Firstly, the relevant theoretical knowledge of data mining and association rule mining is introduced.The basic idea of Apriori algorithm is analyzed and improved and applied to the field of web data mining.Secondly, because without taking into account the importance of the items in the database, it will produce uninteresting rules. As a result, the thinking of item weighting is introduced into association rule mining.Several algorithms and models of weighted association rule mining are studied in depth.The advantages and disadvantages of existing models and algorithms of weighted association rules are analyzed.And the thought of the improved algorithm based on this idea is elaborated.Thirdly, an improved algorithm of weighted association rules based on matrix is proposed.It only needs to scan the transaction database once to convert it into the database in the form of 0-1 matrix, and reduces the memory space occupied. This algorithm pre-prunes frequent (k-1)-itemsets before connecting, and improves the pruning strategy. It does not produce candidate itemsets, but directly generate frequent itemsets.The introduction of the weight cause that the superset of frequent itemsets may be infrequent itemsets. The way to generate frequent 2-itemsets is considered separately so that the weighted frequent itemsets will not be lost. When generating the association rules, it introduces interestingness constraint. The algorithm pseudo-code and flow chart are given, illustrating the feasibility and superiority of the algorithm through an example and an experiment.Finally, the process of personalized recommendation is introduced and the improved algorithm is applied to the field of personalized recommendation for knowledge points. The personalized recommendation includes the offline part and online part, and the main advantage of the improved algorithm is in the offline part. It saves the offline time of generating weighted association rules. Simulated test shows that the system can recommends knowledge points for learners scientifically.