Association Rules Based Background Knowledge Attack And Privacy Protection

For the purposes of government affairs, information sharing, research and the other, government departments, research institutions and other data collectors collected data will be selectively released. Because the information released may be involved in the privacy of individuals, such as prevalence, consumption records, social relations, which require the release of the data privacy at the same time, the attacker can not infer a high degree of confidence the individual's sensitive information specific to to ensure the privacy of information security, which requires anonymity of the original data processing and data dissemination for the purpose of the use and analysis of data, so the data once it has been the anonymity of the release, as well as reduce the availability of existing research work mainly in the protection of individual privacy will not be disclosed at the same time makes the data have a higher availability.Existing data for the release of sensitive information and privacy attacks can be divided into two categories, one is directly inferred according to data released by the now widely used K-anonymity, L diversity of privacy rules to address such attacks; another species with the background of the attacker has mastered the knowledge to infer sensitive information about individual goals, existing methods assume that the data released by the attacker may have some specific background knowledge in privacy model based on the original by adding constraints, part of the solution the background knowledge attack. But data released in a row, the attacker could use a continuous release of large amounts of data mining association rules as a background knowledge of personal privacy attacks, resulting in the privacy of information disclosure, existing methods are not yet targeted discussion.To solve this problem, this paper published data for the continuous background attack. Using association rules to data released by the continuous quasi-identifiers and sensitive attributes the association between attributes, the formation of positive and negative association rules as background knowledge. Associate with the conditional probability that sensitive identity attributes and the links between attributes, based on background knowledge modeling probabilistic method, making target-sensitive property of the association between the individual and the confidence level changes resulting in leakage of privacy; attacker using entropy measure Caused by the background level of privacy disclosure.Published data for the continuous attacks on the background knowledge, this paper presents a new privacy rules (ε,λ)-distinctness and privacy protection algorithms. (ε,λ)-distinctness rules are used to prevent an attacker with a background in the quasi-identifier property speculation after the property and sensitive relationship, allow an attacker to guess based on objective and accurate background knowledge of sensitive property of the individual probability of no more than 1/ε, Not have the background knowledge of the attacks target the individual who suggested the probability of sensitive property does not exceed 1/(ε*λ). Proposed (ε,λ)-distinctness rules algorithm, the use of grouping algorithm, computational satisfy (ε,λ)-distinctness rules anonymous data, and analysis of (ε,λ)-distinctness the security of privacy rules.This selection of commonly used international Adult database from the two experiments verify the privacy protection methods. In resisting the attacks on the background knowledge, analysis and comparison of the (ε,λ)-distinctness rules and rules for the performance of Anatomy, experimental results show that the (ε,λ)-distinctness rules than rules that are more resistant than the Anatomy background knowledge attack. In data availability and computational efficiency, analyzed and compared (ε,λ)-distinctness rules, Anatomy rules and 1-diversity performance, experimental results show that, (ε,λ)-distinctness rules have a higher data availability and efficiency.