t-closeness
t-closeness is a further refinement of l-diversity group based anonymization that is used to preserve privacy in data sets by reducing the granularity of a data representation. This reduction is a trade off that results in some loss of effectiveness of data management or mining algorithms in order to gain some privacy. The t-closeness model extends the l-diversity model by treating the values of an attribute distinctly by taking into account the distribution of data values for that attribute.
Attacks on l-diversity
This is useful because in real data sets attribute values may be skewed or semantically similar. However, accounting for value distributions may cause difficulty in creating feasible l-diverse representations. The l-diversity technique is useful in that it may hinder an attacker leveraging the global distribution of an attribute's data values in order to infer information about sensitive data values. Not every value may exhibit equal sensitivity, for example, a rare positive indicator for a disease may provide more information than a common negative indicator. Because of examples like this, l-diversity may be difficult and unnecessary to achieve when protecting against attribute disclosure. Alternatively, sensitive information leaks may occur because while l-diversity requirement ensures “diversity” of sensitive values in each group, it does not recognize that values may be the semantically close, for example, an attacker could deduce a stomach disease applies to an individual if a sample containing the individual only listed three different stomach diseases.
Formal definition
Given the existence of such attacks where sensitive attributes may be inferred based upon the distribution of values for l-diverse data, the t-closeness method was created to further l-diversity by additionally maintaining the distribution of sensitive fields. The paper t-Closeness: Privacy beyond k-anonymity and l-diversity (2007) defines t-closeness as being:
The t-closeness Principle: An equivalence class is said to have t-closeness if the distance between the distribution of a sensitive attribute in this class and the distribution of the attribute in the whole table is no more than a threshold t. A table is said to have t-closeness if all equivalence classes have t-closeness.
Aggarwal and Yu (2008) state that with this definition, threshold t gives an upper bound on the difference between the distribution of the sensitive attribute values within an anonymized group as compared to the global distribution of values. They also state that for numeric attributes, using t-closeness anonymization is more effective than many other privacy-preserving data mining methods.
See also
References
- Charu C. Aggarwal; Philip S. Yu, eds. (2008). "A General Survey of Privacy". Privacy-Preserving Data Mining – Models and Algorithms (PDF). Springer. ISBN 978-0-387-70991-8.
- Ashwin Machanavajjhala, Daniel Kifer, Johannes Gehrke, and Muthuramakrishnan Venkitasubramaniam (March 2007). "l-Diversity: Privacy Beyond k-Anonymity" (PDF). ICDE (Cornell University). doi:10.1145/1217299.1217302.
Background Knowledge Attack. Alice has a pen-friend named Umeko who is admitted to the same hospital as Bob and whose patient records also appear in the table shown in Figure 2. Alice knows that Umeko is a 21-year-old Japanese female who currently lives in zip code 13068. Based on this information, Alice learns that Umeko’s information is contained in record number 1,2,3, or 4. Without additional information, Alice is not sure whether Umeko caught a virus or has heart disease. However, it is well known that Japanese have an extremely low incidence of heart disease. Therefore Alice concludes with near certainty that Umeko has a viral infection.
- Ninghui Li, Tiancheng Li, and Suresh Venkatasubramanian (2007). "t-Closeness: Privacy beyond k-anonymity and l-diversity" (PDF). ICDE (Purdue University). doi:10.1109/ICDE.2007.367856.