112 G. Pitsilis et al To measure the accuracy we calculated the Mean Absolute Error between the di- ectly calculated similarity s and the one derived from the transitive trust S.We use the following formula in which Cmax and Cmin are the maximum/minimum val ues of the Correlation Coefficient(I and -1 respectively) MAE (5.1) The evaluation algorithm can be described in pseudo-code as in fig. 3. Let us call dt j the direct trust between entities i and j and iti j the indirect one. Assuming that j is within 2 hops of i in the constructed trust graph, the indirect trust of i for j can be calculated using subjective logic in two steps: First, the derived trust of every alter- native path that begins from i and ends to j is calculated separately as a transitive lationship using the suggestion operator & Then all the values of the alternative paths along with dTi,j are combined together using the consensus operator e which gives the value of iTij. In general the consensus is expressed in the following for mula where A and B are two different agents which hold about the statement p re- spectively the opinions @n ando Let be the set of all users Let r be the set of all ratings over items Let R,cR be the set of the ratings of some user u k/m/会D s Cardinality of set of ratings of user i Let ecR The set of ratings of user i Let Mcki: pEM, EpCR and E, nE2 10 For j in M do p has 10 common ratings S←CC(,) Pearson' s similarity李 T←iast(,j Derived Indirect trust s S←f(T) Derived Similarity from our formula s s Absolute mean error value s MAE← End For j End for i Fig. 3. The evaluation algorithm The consensus opinion held by an imaginary agent A, B is o=o2={b2,d2,u (5.2) More about this can be found in[20]. In our particular case the statement p is th trustworthiness of the target j. A, B represent the alternative paths from i to The algorithm for calculating the indirect trust between the origin i and the target j is shown in figure 4To measure the accuracy we calculated the Mean Absolute Error between the directly calculated similarity S and the one derived from the transitive trust S’. We use the following formula in which Cmax and Cmin are the maximum/minimum values of the Correlation Coefficient (1 and -1 respectively): max min ' CC SS MAE (5.1) The evaluation algorithm can be described in pseudo-code as in fig. 3. Let us call dti,,j the direct trust between entities i and j and iti,,j the indirect one. Assuming that j is within 2 hops of i in the constructed trust graph, the indirect trust of i for j can be calculated using subjective logic in two steps: First, the derived trust of every alternative path that begins from i and ends to j is calculated separately as a transitive relationship using the suggestion operator
. Then all the values of the alternative paths along with dTi,,j are combined together using the consensus operator which gives the value of iTi,,j. In general the consensus is expressed in the following formula where A and B are two different agents which hold about the statement p respectively the opinions A Z p and B Z p . Fig. 3. The evaluation algorithm The consensus opinion held by an imaginary agent A,B is: },,{ , ,,, BA p BA p BA p B p A p BA p ZZZ udb (5.2) More about this can be found in [20]. In our particular case the statement p is the trustworthiness of the target j. A, B represent the alternative paths from i to j. The algorithm for calculating the indirect trust between the origin i and the target j is shown in figure 4. Let K be the set of all users Let R be the set of all ratings over items Let RuR be the set of the ratings of some user u Let KiK : Ru t10 * Cardinality of set of ratings of user i * For i in Ki Let Ei R * The set of ratings of user i * Let MKi : pM , EpR and EE pi t 10 For j in M do * p has 10 common ratings with i m CCS ji ),( * Pearson’s similarity * m iTrustT ji ),( * Derived Indirect trust * )( ' m TfS * Derived Similarity from our formula f * max min ' CC SS MAE m * Absolute Mean Error value * End For j End For i Average(MAE) 112 G. Pitsilis et al