Pipeline of dissimilarity calculation (taking the data fragment in the lower part of Table I as an example). In Step 1, we first normalize the numerical attribute A1 and then construct graphs for the three attributes according to Remark 4. Since values of A1 are normalized into [0, 1], which yields theta =1, we set theta =1 for all the graphs. Steps 2 and 3 are demonstrated by taking the computation of Psi3(o31,o32) as an example. In Step 2, we first prepare u311=[{0,1,0,0,0}], u312=[{0.33,0,0.33,0.33,0}], and u313=[{0,0,0,0,1}], then we successively compute psi31(o31,o32) and w31 according to (6) and (7). In the same way, we obtain psi32(o31,o32), w32, psi33(o31,o32), and w33. Finally, in Step 3, the subdissimilarities and the corresponding weights are combined according to (5) to form Psi3(o31,o32).
