A Differential Privacy Framework for Collaborative Filtering

Algorithm 1

The perturbed Pcc (or Cos) algorithm.

Input: the rating matrix, the privacy budget , the parameter k

Output: the predicted scores

.

2. simMat = similarity (activeMatTrain, otherMatTrain)

3. simMat = simMat + Lap

4. otherAvgVec = mean () + Lap

5. otherMatPred = otherMatPred - otherMatAvg

6. for each active user j

7.activeAvg = mean (activeMatTrain (j)) + Lap

// The spdiags function transforms the similarity vector of user j into a diagonal matrix.

8.activeMatPred (j) = sum (spdiags (simMat (j))otherMatPred) + activeAvg

9. end for

10. return activeMatPred