Research On Privacy Preserving Methods For Outsourcing Services Of Association Rule Mining

In today’s society, with the technological advancements，information technologyand database technology has been rapidly expanded, many companies or organizationshave accumulated a large amount of data. How to get useful information from thesemassive data has aroused widespread attentions. The ultimate goal of data mining is togain some valuable information and knowledge hidden in these massive data. Datamining has increasingly played an obvious and important role in business decisions,market analysis and other fields. Data mining can bring great benefits, but it needs veryhigh requirements on software and hardware environment. Outsourcing this work toprofessional services is a quite good choice for the enterprises or organizations whichare lack of resources. However, it will inevitably involve security information andprivacy issues. Therefore, it is necessary to transform the original data beforeoutsourcing it to professional services. Only by so doing, this treatment can providebetter protection for the privacy. This thesis mainly studies the problem of privacyprotection in outsourcing services of association rule mining, related work include:①Privacy protection technology in general association rule mining are studied,which includes summarizing of several typical methods and analysing their advantagesand weaknesses.②Privacy preserving technology for association rules mining in an outsourcingenvironment has been studied. Its basic requirements are given. Otherwise the thesisdescribes the similarities and differences between outsourcing and non-outsourcingenvironment, gives a basic process of outsourcing privacy algorithm for MiningAssociation rules.③This thesis learns several existing privacy protection method in outsourcingassociation rule mining and analyses their strengths and weaknesses. Twoimprovements are presented for Bloom Filters-based outsourcing association rulesmining algorithm.The first improvement: The original transaction database is converted into theform of a Bloom filter using independent mapping space Bloom filter and reversibleaddress mapping function. Then to compress transaction in accordance with thetransformed each transaction records Hamming weight vector, and do ’ and ’ operationusing matrix column vectors, calculate the support of candidate frequent itemsets, and ultimately get the frequent itemsets. Experimental results show that the improvedmethod not only has good reversibility, but also increase the efficiency significantly.The second improvement: the Bloom filter and hash irreversible address mappingfunction are also used, the difference is rearranging the mapping table which is neededin restore process. This method greatly reduces the storage space of the algorithm, andreduces the overhead.④Lastly verified through experiments using simulated data and real data sets ontwo improved methods. Experimental results show that both two improved methodswere able to achieve good results, can effectively apply on the actual.