Study And Improvement On K-Anonymity Model For Private Protection

With the rapid development of information society, people share more and more information. Free access micro-data publication is an effective means for scientific research and information interchange and also provides a platform for data sharing. However, with the application of the data-mining technology and the power of Search Engines, the private protection of micro-data publication becomes worse and the outcry of private protection becomes stronger. It not only protects private ineffective but also makes data useless in using old methods to solve this problem. K-anonymity model is one of the effective models for its simple ideology and easy implementing, which is a better choice for private protection in micro-data publication. Just as every coin has its two sides; K-anonymity model also has its limitation. In this paper, after making an intensive study of K-anonymity, it points the limitation of the model, and puts forth a new model which can deal with the limitation of K-anonymity. The specific content as follows:①On the basis of former research and existent solutions, this article studies the use of K-anonymity in private protection detailedly, and focuses on the limitation of it. At the same time, it reveals the issue of attribute privation degree of attribute disclosure through further analysis. Different attribute value can be given different attribute protection degree basing on their importance, which can decrease information loss.②Through analysis of the drawbacks of K-anonymity model, which is a popular model currently, this article puts forth a new model for private protection--(alp,dif)-anonymity model. This model can not only deal well with attribute disclosure,which is a difficult problem for K-anonymity model, but also take attribute protection into account. All this can ensure quasi-identifier attributes lose less information loss. To make sure the reliability of the publishing data, the new mode adopts generalization to preprocess it, and puts forth a new greedy algorithm bases on information loss. It takes both information loss and attribute protection degree into account to choose attributes that need to be generalized.③Last part of the article is the experiments which use the classical dataset. Experiments show that the new model can deal with the attribute disclosure problem as well as ensure the reliability of the publishing data, and not consume too much time at the same time.