Experimental Metrics
One of the statistical prediction accuracy metrics for evaluating recommender systems is MAE(Mean Absolute Error) which is the mean of the errors of the actual user ratings against the predicted ratings in individual prediction[2][7][9]. MAE is computed by Equation (2). In the equation, N is the total number of predictions and is the error between the predicted rating and the actual rating. The lower MAE is, the more accurate prediction with respect to the numerical ratings of users we get.
. (2)

Experimental Results
The proposed recommendation algorithm(KMGCF) is compared with four other methods. The first method is the pure collaborative filtering(CF) method used by GroupLens[3][4]. The second method is the method that collaborative filtering is applied after the KNearest Neighbor method selected the neighbors. We call it KNCF. The third one is the collaborative filtering method with the KMeans clustering. We denote it as KMCF. The last one is the collaborative filtering method with the graph approach. We call it GCF.
The experimental results are shown in Table 1. We determined 100 as the optimal value of k for KNCF and 21 as the optimal value of k for KMGCF through various experiments. The numbers within the parenthesis next to GCF and KMGCF indicate the depth to search in the graph. Because we obtained the optimal number k of clusters through various experiments, K=21 has been used for KMGCF. In the parameters of GCF and KMGCF, L and H denote that the threshold values for the Pearson correlation coefficients. For example, if L = 0.8 and H = 0.7 means that we’ll select the neighbors whose similarity is less than –0.8 and also larger than 0.7.
Table 1. Comparison of various methods
Share with your friends: 