Secure Personalized Recommendation System for Mobile User - Information Security and Cryptology-ICISC 2010 - page 272

although there is a higher degree of the similarity to the target user. It follows that
users who have rated a large number of items will belong to more neighborhoods than
those users who have rated few items. This is a potential security risk in the context of
profile injection attacks.

Fig. 3. Recommendation Process by Modified RPCF Algorithm for Hotel Enquiry

The significance weight of a target user u for a neighbor v is computed as:

⎧ n

sim * n < N

⎪ ⎨

W =

u , v

(1)

N

u , v

⎪ ⎩ sim otherwise

u , v

where n is the number of co-rated items, N is a global constant, and sim u,v is Pearson's
correlation coefficient. A prediction for the target user is computed by using equation
(2), replacing sim u,v with w u,v .
An attack profile with a very large number of filler items will necessarily be
included in more neighborhoods, regardless of the rating value. The risk can minimize
because a large filler size threshold is required to make the attack successful. In most
cases, genuine users rate only a small portion of all recommendable items; therefore,
an attack profile with a very large filler size is easier to detect [17].

m

∑

sim ( r − r )

v

uv , vi ,

Pr =+

i = 1

(2)

u

ui ,

m

∑

sim

uv ,

i

i = 1

The profile injection attacks are detected by applying modified RPCF algorithm.
Trust model can improve in collaborative filtering [12]. By explicitly calculating a