CTKM: Crypto-Based User Clustering on Web Transaction Data.

User transaction data are rich, valuable, but sensitive. With the huge amounts of transaction data, data mining algorithms can make many applications practical, such as customer-behavior analysis, marketing, and forensics. The value behind the transaction data analysis on the other hand raises the risk of data leak. In this paper, we introduce a C rypto-based KM eans clustering algorithm ( CTKM ) on the T ransaction data of web users for user clustering and data protection as well. Considering the categoricalness of user transaction data, a taxonomy-based distance has been employed, which is applicable to the data encryption process also. In order to obtain efficient computations on the distance, a distance batch computing ( DBC ) protocol is designed and deployed in a two-server platform. We theoretically estimate both the computation and communication costs of the algorithm. Experimental results on a real data set demonstrate its practical value on web user clustering.