15], which surveys model-based attack resistant algorithms and proposes a robust
matrix factorisation strategy.
A model-based recommendation strategy based on clustering user profiles is
analysed in [21]. In this strategy, similar users are clustered into segments and the
similarity between the target user and a user segment is calculated. For each segment, an aggregate profile, consisting of the average rating for each item in the
segment is computed and predictions are made using the aggregate profile rather
than individual profiles. To make a recommendation for a target user u and target
item i, a neighbourhood of user segments that have a rating for i and whose aggregate profile is most similar to u is chosen. A prediction for item i is made using
the k nearest segments and associated aggregate profiles, rather than the k nearest
neighbours. Both k-means clustering and PLSA-based clustering, as described in
Section 25.5.3.2, are evaluated. The prediction shift achieved by an average attack
on these algorithms, compared with the standard kNN algorithm, is shown in Figure 25.11 (left). The model-based algorithms are considerably more robust and not
significantly less accurate, since, according to [21], PLSA and k-means clustering
achieve an MAE of 0.75 and 0.76 using 30 segments, in comparison to a value of
0.74 for kNN.