From fig 3 we can see that the mae and rmse values

Info icon This preview shows pages 16–19. Sign up to view the full content.

From Fig. 3 , we can see that the MAE and RMSE values ¯rst decrease, then reach a threshold ( ² ¼ 0 : 4) that achieves the lowest estimation errors, and further increase along with the increase of ² . Remembering Eq. ( 13 ) where ² is the proportion of user Table 4. Performance comparison (a smaller value means better performance). TD TD ¼ 5% TD ¼ 10% TD ¼ 15% TD ¼ 20% Model MAE RMSE MAE RMSE MAE RMSE MAE RMSE UserMean 0.8782 1.8586 0.8772 1.8561 0.8766 1.8537 0.8769 1.8561 ItemMean 0.7665 1.5776 0.7281 1.5485 0.6863 1.5306 0.6795 1.5308 UPCC 0.7579 1.5342 0.7144 1.4941 0.6319 1.4464 0.5927 1.4232 IPCC 0.7183 1.5145 0.7352 1.5076 0.6998 1.4782 0.6507 1.4473 WSRec 0.7632 1.5360 0.6806 1.4442 0.6337 1.4047 0.6120 1.3864 MF 0.5793 1.5099 0.5683 1.3947 0.5438 1.3726 0.5328 1.3576 NLBR 0.5520 1.3953 0.5254 1.3376 0.5073 1.3245 0.4857 1.3067 RAP 0.5402 1.3816 0.5247 1.3349 0.5165 1.3183 0.4796 1.3013 ULMF 0.5351 1.3714 0.5134 1.3143 0.5084 1.3107 0.4762 1.2937 SLMF 0.5421 1.3767 0.5189 1.3243 0.5013 1.3049 0.4742 1.2887 JLMF 0.5393 1.3746 0.5167 1.3203 0.5041 1.3072 0.4750 1.2907 626 Y. Yin et al. Int. J. Soft. Eng. Knowl. Eng. 2016.26:611-632. Downloaded from by WEIZMANN INSTITUTE OF SCIENCE on 07/01/16. For personal use only.
Image of page 16

Info icon This preview has intentionally blurred sections. Sign up to view the full version.

location-aware model and ð 1 ³ ² Þ is the proportion of SLMF. ² ¼ 0 : 4 illustrates that in the ¯nal prediction result, the result of SLMF should take a more important role. Also, our model gives enough room to choose a suitable ² (0.3 0.6) to achieve a superior prediction accuracy than baselines (see Table 3 ). 5.6. The sensitivity to ¸ The parameter ! is the regularization parameter in the MF-based models (see Eqs. ( 2 ), ( 10 ) and ( 12 )). The parameter ! controls the impact of the regularization terms on the objective function. If ! is too small, our proposed models will focus on the MF part, but ignore the function of network location-aware regularization terms. In contrast, if ! is too large, the location information are likely to weaken the prediction performance of the MF part. In this section, we study the sensitivity of our proposed model JLMF to ! under the default parameter setting, and the training set densities are 10%, 15% and 20%. The experimental results are shown in Fig. 4 . In Fig. 4 , we can observe that the MAE and RMSE values achieve the smallest values at ! ¼ 0 : 01 in all cases of training set densities. It indicates that ! should be set to a moderate value. A too small value (for example, ! ¼ 0 : 0001) or a too large value (for example 1) cannot properly balance the proportions of the MF part and regularization part. In this paper, we set ! ¼ 0 : 01 as the default value. (a) (b) (c) (d) Fig. 3. The sensitivity to ² . (a) Training Matrix Density = 10%. (b) Training Matrix Density = 10%. (c) Training Matrix Density = 15%. (d) Training Matrix Density = 15%. QoS Prediction for Web Service Recommendation with Network Location-Aware Neighbor Selection 627 Int. J. Soft. Eng. Knowl. Eng. 2016.26:611-632. Downloaded from by WEIZMANN INSTITUTE OF SCIENCE on 07/01/16. For personal use only.
Image of page 17
5.7. The sensitivity to ° The parameter ± is the learning rate in the gradient descent algorithm (see Eq. ( 5 ), in Sec. 4.2 ). If ± is too small, the algorithm will need more iteration times to achieve the local optimal. In contrast, if ± is too large, it is possible for the algorithm to be hard to converge. In this section, we study the sensitivity of our proposed model to ± under the default parameter setting, and the training set densities are 10%, 15% and 20%.
Image of page 18

Info icon This preview has intentionally blurred sections. Sign up to view the full version.

Image of page 19
This is the end of the preview. Sign up to access the rest of the document.
  • Winter '15
  • MAhmoudali
  • Analysis of algorithms, Computational complexity theory, Service system, Expectation-maximization algorithm, Weizmann Institute of Science

{[ snackBarMessage ]}

What students are saying

  • Left Quote Icon

    As a current student on this bumpy collegiate pathway, I stumbled upon Course Hero, where I can find study resources for nearly all my courses, get online help from tutors 24/7, and even share my old projects, papers, and lecture notes with other students.

    Student Picture

    Kiran Temple University Fox School of Business ‘17, Course Hero Intern

  • Left Quote Icon

    I cannot even describe how much Course Hero helped me this summer. It’s truly become something I can always rely on and help me. In the end, I was not only able to survive summer classes, but I was able to thrive thanks to Course Hero.

    Student Picture

    Dana University of Pennsylvania ‘17, Course Hero Intern

  • Left Quote Icon

    The ability to access any university’s resources through Course Hero proved invaluable in my case. I was behind on Tulane coursework and actually used UCLA’s materials to help me move forward and get everything together on time.

    Student Picture

    Jill Tulane University ‘16, Course Hero Intern