Journal of Convergence Information Technology Volume 5. Number 8 October 2010 shows the influence of population size on the performance of the recommender system. As expected, the performance improves as the population size increases, but it reaches a saturation point at PS=50 and any further increment does not improve results. Accordingly, we will fix the population size to 50 while evaluating the performance of various approaches Recall +Precision 10 Figure 3. Performance results of GAWMCF approach vs population size 5.3. 1. Performance comparison Recalling that the CF algorithm uses a nearest-neighbor model for generating predictions, a crucial step in CF is the selection of a neighborhood. The number of nearest neighbors used for neighborhood formation can substantially affect the system's accuracy. However, the optimal size of a neighborhood usually depends on the nature of the problem. a pilot experiment of this study has been conducted and suggested 30 as the optimal value. To evaluate the sensitivity of different recommendation numbers, we fix the number of eighbors to 30 and perform an experiment with different recommendation numbers of 3, 5, 10 20, 25, and 30. The comparisons of precision, recall, and Fl among various approaches are shown from Fig 4 to Fig. 6 As expected, when the number of recommendations increases, the precision drops smoothly but the recall improves gradually. In all cases, the single-criterion approach SUCF performs worse than the other three multi-criteria approaches. Among the three multi-criteria approaches, GAWMCF performs the best, followed by GAFMCF, and then EWMCF. The results demonstrate the effectiveness of the GA-based feature weighting and confirm the previous studies in that the wrapper-based approach outperforms the filter-based approach. Finally according to Figure 6, we can see that the Fl-measure reaches an optimal value for RN aroune 0.7 A-EWMCF SUCE 351015202530 Figure 4. Precision-metric for various approaches vs recommendation numberJournal of Convergence Information Technology Volume 5, Number 8, October 2010 shows the influence of population size on the performance of the recommender system. As expected, the performance improves as the population size increases, but it reaches a saturation point at PS = 50 and any further increment does not improve results. Accordingly, we will fix the population size to 50 while evaluating the performance of various approaches. 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 20 30 50 75 100 Performance Metric PS F1 Recall Precision Figure 3. Performance results of GAWMCF approach vs. population size 5. 3. 1. Performance comparison Recalling that the CF algorithm uses a nearest-neighbor model for generating predictions, a crucial step in CF is the selection of a neighborhood. The number of nearest neighbors used for neighborhood formation can substantially affect the system’s accuracy. However, the optimal size of a neighborhood usually depends on the nature of the problem. A pilot experiment of this study has been conducted and suggested 30 as the optimal value. To evaluate the sensitivity of different recommendation numbers, we fix the number of neighbors to 30 and perform an experiment with different recommendation numbers of 3, 5, 10, 15, 20, 25, and 30. The comparisons of precision, recall, and F1 among various approaches are shown from Fig. 4 to Fig. 6. As expected, when the number of recommendations increases, the precision drops smoothly but the recall improves gradually. In all cases, the single-criterion approach SUCF performs worse than the other three multi-criteria approaches. Among the three multi-criteria approaches, GAWMCF performs the best, followed by GAFMCF, and then EWMCF. The results demonstrate the effectiveness of the GA-based feature weighting and confirm the previous studies in that the wrapper-based approach outperforms the filter-based approach. Finally, according to Figure 6, we can see that the F1-measure reaches an optimal value for RN around 20. 0.1 0.2 0.3 0.4 0.5 0.6 0.7 3 5 10 15 20 25 30 Precision RN GAWMCF GAFMCF EWMCF SUCF Figure 4. Precision-metric for various approaches vs. recommendation number 133