Privacy-Preserving Collaborative Data Mining:
2008-2-6 2 Privacy-Preserving Collaborative Data Mining Data Mining Data Set A Data Set B Results Data Set C Alice Carol Bob
Problem Definition:
2008-2-6 3 Problem Definition Goal: Multiple parties jointly conduct sequential pattern mining without revealing their private data to each other.
An example: Pattern: ATM < ticket < pop-corn with support of 1/2 c1 c2
Approach:
2008-2-6 4 Approach Support: a sequential pattern (x < y) has support s% if s% of transactions (records) in a joint data set of n parties contain both x and y with x happening before y.
Approach:
Construct event vectors (Col: transaction times of an item; Row: customer-ID) from data tables.
Compute the support of sequential patterns (or events) based on event vectors via a secure protocol.