HI gh-imensional Data

Existing research onprivacy-preserving datapub- lishing focuses onrelational data:inthiscontext, theobjec- tiveistoenforce privacy-preserving paradigms, suchask- anonymity andf-diversity, while minimizing theinformation loss incurred intheanonymizing process (i.e. maximize datautility). However, existing techniques adoptanindexing- orclustering- basedapproach, andworkwellforfixed-schema data,with lowdimensionality. Nevertheless, certain applications require privacy-preserving publishing oftransaction data(orbasket data), whichinvolves hundreds oreventhousands ofdimensions, rendering existing methods unusable. We propose anovelanonymization methodforsparse high- dimensional data.We employa particular representation that captures thecorrelation intheunderlying data, andfacilitates theformation ofanonymized groupswithlowinformation loss. We propose anefficient anonymization algorithm basedonthis representation. We showexperimentally, using real-life datasets, thatourmethodclearly outperforms existing state-of-the-art in termsofbothdatautility andcomputational overhead.