Anonymizing Binary Tables is APX-hard

The problem of publishing personal data without giving up privacy is increasingly important. An interesting formalization is the k-anonymization, where all rows in a table are clustered in sets of at least k records, and all the entries for which records in the same cluster have different values are suppressed. The problem has been shown to be NP-hard when the records values are over a ternary alphabet and k = 3. In this paper we show that the problem is not only NP-hard, but also APX-hard, when the records values are over a binary alphabet and k = 3.