Research On Micro-aggregation Algorithm For Personalized Privacy Protection

In recent years, the data released in the scientific research and data analysis has been widely concerned. Due to the data sets to be released often contain sensitive information, once the leakage happened, there will be a huge risk of leakage. Therefore, the main problem faced is how to ensure the data utility and privacy preservation, in the publication of datasets. Anonymous is a very classical method, the main idea is:the quasi identifier attributes are anonymous based on removing the unique identifier attribute, that records within the anonymous group cannot be divided. At present, there are many kinds of anonymous models, which are represented by L diversity, (a, k) anonymous model and so on. However, most of those models focus on protecting privacy exerting identical protection for the whole table with pre-defined parameters. As a result, it could not meet the diverse requirements of protection degrees varied with different sensitive values.To sum up, this paper presents a model-the alpha diversity k-anonymity, which takes two aspects of a global perspective and the personalized requirements into consideration, to meet the unrelated constraint of diverse sensitive information. There are two ways to implement the k anonymity, generalization/suppression and micro-aggregation. The algorithm of generalization/suppression is proved to be existing problems of low efficiency and poor data utility, micro-aggregation is a good choice. Because of the traditional micro-aggregation algorithm cannot meet the demand of the anonymous model proposed in this paper, we design an improved algorithm framework based on traditional micro-aggregation algorithm, to implement the anonymity model.By using the framework proposed in this paper to implement the anonymity model, it can not only be easy to implement, but also improve the efficiency of the data and reduce the risk of privacy disclosure. To verify the validity of the scheme, we compare our model with other two models:k-anonymity and (0.5,k)-anonymity, and conduct several experiments in real data sets, and analyze the results from the data validity and privacy disclosure risk and time complex degree. Experimental results show that, our scheme is mostly superior to other models in the data sets.