【作者】 王春丽；

【导师】 柴玉梅；

【作者基本信息】 郑州大学 ， 计算机软件与理论， 2011， 硕士

【摘要】 20世纪80年代以来,随着数据库和信息技术的发展与广泛应用,各个应用领域均累积了海量的数据,且数据呈几何级数增长。而存储在大型数据库中的海量数据包含了大量的、有利于决策者作出英明判断的信息。但是由于缺乏强有力的数据分析工具,这些数据就变成了无人问津的“数据坟墓”。这样,基于直觉和经验的决策者往往不能做出有利于应用领域的决定,使得这些包含有价值的数据无法发挥它们应有的决策指导作用。如何从海量数据集中提取出有利于商务决策的信息成为各个领域的需求,在这种情况下,数据挖掘应运而生,从不同的角度以不同的形式为商务策略、科学等各领域提供依据,推动了信息技术的迅猛发展。本文研究的两项工作：首先,频繁项集挖掘作为关联规则产生的首要步骤,其挖掘效率的高低直接影响着关联规则产生的总体效率。而数据集扫描次数和支持数计算效率一直是影响频繁项集挖掘的主要因素。本文以shariq bashir提出的TOP-N闭频繁项挖掘算法为基础,利用位向量映射技术,提出了基于索引的频繁项集挖掘算法。该算法用位向量表示数据集,仅需一次扫描数据集。频繁项集的挖掘过程采用集合枚举树的节点构造过程来表示,在创建节点枚举项的尾项支持数时,利用枚举项集的位索引,计算枚举项合并形成的新项集的支持数,同时引入广度扩展剪枝和区域索引跳过策略,从而快速地挖掘出频繁项,有效地提高了算法的执行效率；实验结果表明,该算法在稀疏数据集上有很高的运行效率。其次,由于挖掘出的频繁项集可能包含大量的用户不感兴趣的规则,同时如果没有进一步的分析或领域知识,频繁项集并不能直接用于预测。如何消除频繁项集中用户不感兴趣的频繁项是频繁项集用于指导实践的瓶颈。所以,本文在基于索引的频繁项集挖掘算法基础上,本文又提出基于频繁项集的互补替代关系挖掘算法。该算法通过对已挖掘出的各频繁项集中的频繁项进行相关性计算,挖掘出频繁项之间的互补替代关系,去除噪声数据的影响,以图形的方式显式地向决策者展示出各个频繁项之间的互补(或替代)性,更便于决策者做出准确、合理的判断。实验表明,CAG比频繁项集表示的信息更有效,更精确。更多还原

【Abstract】 The database and the information’technology have highly developing and widespread applicating in many domain Since the 1980s, a large number of historical data have not only accumulated in real life and commercial applications,but aslo increase in Geometric progression’s way.The data Storing in the large database system contain mass of information which are advantageous for the decision makers to make astute decision.Due to lacking a powerful data analytical application,these data became "the data’grave" that many people weren’t interested to mention. The dexision maker always make a decison basing on their intuition and experiences.Thus the decisions usually incapability in applied realm.These data contain many worthy informatios and resources,and they should guide the decision-maker in providing an implicit information. But,they can not act well in fact.How to mine informations from the large database becomes the demand for many applied realm. In this case, the data mining technology emerges with a new subject.It provides a basis by many different ways for many domain,such as commerce strategy,science,and so on, and makes the information technology fast development.Frequent item sets mining is the first step in generating association rules,its mining efficiency is directly related to the efficiency of generating association rules,But It has been expanded by the number of scanning dataset and efficiency of computing support in mining frequent itemsets. Based the algorithm for ming Top-N closed frequent itemsets, proposed by shariq bashir,this article first proposed an algorithm for mining frequent itemsets. It only need to scan dataset once using bit-vector to represent dataset.The process of mining frequent itemsets is represented by the process of creating the node of enumeration tree.While it is creating the tailset of the enumeration itemsets, using PBR-Index of the enumeration itemsets of head,the algorithm computes the support of the new itemsets combined by the head itemsets and one item in the tailset of the node. Simultaneously,we introduce the breadth expansion pruning strategy and region-index skiming strategy.Thus it can fastly find out all of the frequent itemsets,and effectively improve the the algorithm’s efficiency.Because the frequent itemsets might include a great deal of rule that consumer were not interested,and if there is no further analysis or realm knowledge, the association rule can not apply directly to Prediction.So how to eliminate the rule that consumers is not interested in the frequent pattern is also the bottleneck in mining frequent pattern. Based the first algorithm for mining frequent itemsets,we also proposed another algorithm for mining the frequent itemsets’s correlationship. It computes the relevant computation between the items including in frequent itemsets to find their correlationship.and eliminates the influence of the noise datas, demonstrates their complementary- alternative relationship to the policy-maker by CAG,so it is advantageous for the policy-maker to make reasonable and accurate judgment. Results suggested that the algorithm for mining complement-alternative relationship in this article is more efficient and more precise than frequent itemsets in expressing informations.更多还原