【作者】 孙俊；

【导师】 张曦煌；

【作者基本信息】 江南大学 ， 计算机科学与技术， 2016， 硕士

【摘要】 随着科学技术的发展,数据的快速增长,人们迫切需要一种技术来处理这些数据。数据挖掘就是在这样的背景下产生的。数据挖掘是发现隐含在数据中有价值信息的理论与方法,它是人工智能和机器学习领域最受关注、最前沿的研究方向之一。关联规则是数据挖掘重要组成部分。频繁项集挖掘又是关联规则挖掘过程的核心,如何有效地挖掘频繁项集一直以来是研究人员关注的热点。但是在实际应用中,爆炸式数据增长和大量频繁项,阻碍了频繁项集广泛应用。因此,如何对频繁项集挖掘算法优化和对频繁项集压缩成为当前研究的重要方向。本文首先介绍了关联规则和频繁项集的相关概念及几种频繁项集挖掘算法,随后详细介绍了频繁项集压缩技术,并简要分析了几种有效的频繁项集压缩算法,最后,本文提出了两个基于节点集频繁项集挖掘算法,基于节点集Top-k频繁项集挖掘算法TBN算法和基于节点集Top-k闭频繁项集挖掘算法TCBN算法。(1)提出了基于节点集的Top-k频繁项集挖掘算法,该算法运用节点集的概念,通过维持一张Top-k-rank表,摒弃了用户对最小支持度的干预,使最小支持度动态化,只需两次遍历数据库,即可生成所需要挖掘的Top-k频繁项集。通过对比实验证明该算法性能优于当前主流算法。(2)提出了基于节点集的Top-k闭频繁项集挖掘算法,该算法继续沿用了节点集的概念,利用其数据结构POC-Tree的特性和闭合项集的性质,提出两个基于POC-Tree快速挖掘闭频繁项集的策略,并通过Top-k-rank表减少候选项集数量,结合两种闭合检测方法保证挖掘结果具有闭合性。TCBN算法通过对比实验证明了该算法在时间和空间上的优越性。通过对本文所提算法进行广泛对比试验,实验结果表明,TBN算法和TCBN算法比当前主流算法性能更加优越,尤其是在稠密数据集上,优势更加突出。所提算法解决了频繁项集挖掘算法在实际应用中结果集规模数量巨大和最小支持度难以设定的难题,并提高了算法性能。这些研究成果在频繁项集实际问题中的应用提供了一种有效的解决问题的途径。更多还原

【Abstract】 With the development of science and technology, the rapid growth of data, there is an urgent need for a method to deal with the data. Data mining comes into being in such a background and it is the theory and method that mining valuable information from the implicit data.Date Mining is one of the most famouse and cutting-edge researches of artificial intelligence and machine learning. Association rule is the most important part of data mining.In the process of mining association rules, frequent itemsets mining is the core of the mining process. How to effectively mining frequent itemsets has long been a focus of attention of researchers. In practice, however, the explosive growth of data and the huge number of frequently itemsets hinder the wide use of frequent item sets. Therefore, how to optimize the algorithm and compress frequent itemsets has become an important direction of current research.This paper firstly introduces the concept and basic mining algorithms about association rules and frequent itemsets, then introduces the related technologies of itemsets compression in detail, and analyzes several effective compression algorithms of frequent itemsets briefly.Finally, this paper proposes two algorithms based on nodesets for mining Top-k frequent itemsets,the TBN algorithm and the TCBN algorithm.(1) This paper propose the TBN algorithm.The TBN algorithm is based on nodesets,using a Top-k-rank table, abandoned artificial for the minimum support intervention, so that the minimum support is dynamic. only two traversal database operations, the use of generating the Top-k mining method for mining frequent itemsets. Analysis and experimental comparison shows that the TBN algorithm consumes less time and space in different degree for different dataset.(2) This paper propose the TCBN algorithm for mining Top-k frequent closed itemsets.The TCBN algorithm presents two strategies for quickly mining frequent closed itemsets,which are on the basis of the new data structure of POC-Tree and the nature of closed itemsets,and reduce the number of candidate by combining two methods for closed detection to ensure that the results are all closed by Top-k-rank table. Analysis and experimental comparison shows that the superiority of the TCBN algorithm in terms of time and space.This paper has carried on the contrast experiment wide of the proposed algorithms which have better performance, especially on dense datasets. The proposed algorithm aims to solve the frequent item sets mining algorithm in practical application. The huge size of item sets and the number of results is difficult to set the minimum support problem, these researches about mining frequent itemsets provide an effective solution to the practical problems.更多还原