Limits...
A network centrality method for the rating problem.

Li Y, Pin P, Wu C - PLoS ONE (2015)

Bottom Line: Our approach is based on the network relations induced between items by the rating activity of the users.Our method correlates better than the simple average with respect to the original rankings of the users, and besides, it is computationally more efficient than other methods proposed in the literature.Moreover, our method is able to discount the information that would be obtained adding to the system additional users with a systematically biased rating activity.

View Article: PubMed Central - PubMed

Affiliation: School of Business Administration, Northeastern University, Shenyang 110819, P.R.China.

ABSTRACT
We propose a new method for aggregating the information of multiple users rating multiple items. Our approach is based on the network relations induced between items by the rating activity of the users. Our method correlates better than the simple average with respect to the original rankings of the users, and besides, it is computationally more efficient than other methods proposed in the literature. Moreover, our method is able to discount the information that would be obtained adding to the system additional users with a systematically biased rating activity.

Show MeSH
Average outcome of the Kendall’s tau measure on the simulations, with 9 different methods, p varying from 0.4 to 1, N = 50 and M = 50.Lower plot shows Student’s t comparison of the NC method with β = 2/(λ+ + N) with respect to average and SD—confidence intervals are shown in the legend.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4382193&req=5

pone.0120247.g006: Average outcome of the Kendall’s tau measure on the simulations, with 9 different methods, p varying from 0.4 to 1, N = 50 and M = 50.Lower plot shows Student’s t comparison of the NC method with β = 2/(λ+ + N) with respect to average and SD—confidence intervals are shown in the legend.

Mentions: Results of the average outcomes for the four cases are reported in the upper parts of Figs. 3, 4, 5 and 6. From these average trends, it comes out clearly out that SD is the best performing measure (but it is more costly than the NC method both in terms of computational time and memory storage), and that averaging is the worse, while the centrality measure with different values of β lies in–between. The best value of β seems to be 2/(λ+ + N). However, we need to take variance into account when analyzing these results. In the Supporting Information, Figures S1 to S4 display the boxplots of all the 200 realizations, for each of the four cases, of the following three measures: DS, average, and NC measure with β = 2/(λ+ + N). The lower parts of Figs. 3 to 6 take also variance into account and plot the Student’s t–test to check whether the centrality measure with β = 2/(λ+ + N) is statistically different from the SD measure and the simple average, as p changes. When (N, M) = (10, 10), (N, M) = (50, 10) and (N, M) = (50, 50) (Figs. 3, 5 and 6) the NC measure is not statistically different from the other two measures for most values of p: it performs significantly better than the average for high p, and significantly worse than the MS measure for low p. But when (N, M) = (10, 50) (Fig. 4), the NC method is always better than the average with 99% statistical confidence, while it is not statistically different form the MS measure for p above 0.6.


A network centrality method for the rating problem.

Li Y, Pin P, Wu C - PLoS ONE (2015)

Average outcome of the Kendall’s tau measure on the simulations, with 9 different methods, p varying from 0.4 to 1, N = 50 and M = 50.Lower plot shows Student’s t comparison of the NC method with β = 2/(λ+ + N) with respect to average and SD—confidence intervals are shown in the legend.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4382193&req=5

pone.0120247.g006: Average outcome of the Kendall’s tau measure on the simulations, with 9 different methods, p varying from 0.4 to 1, N = 50 and M = 50.Lower plot shows Student’s t comparison of the NC method with β = 2/(λ+ + N) with respect to average and SD—confidence intervals are shown in the legend.
Mentions: Results of the average outcomes for the four cases are reported in the upper parts of Figs. 3, 4, 5 and 6. From these average trends, it comes out clearly out that SD is the best performing measure (but it is more costly than the NC method both in terms of computational time and memory storage), and that averaging is the worse, while the centrality measure with different values of β lies in–between. The best value of β seems to be 2/(λ+ + N). However, we need to take variance into account when analyzing these results. In the Supporting Information, Figures S1 to S4 display the boxplots of all the 200 realizations, for each of the four cases, of the following three measures: DS, average, and NC measure with β = 2/(λ+ + N). The lower parts of Figs. 3 to 6 take also variance into account and plot the Student’s t–test to check whether the centrality measure with β = 2/(λ+ + N) is statistically different from the SD measure and the simple average, as p changes. When (N, M) = (10, 10), (N, M) = (50, 10) and (N, M) = (50, 50) (Figs. 3, 5 and 6) the NC measure is not statistically different from the other two measures for most values of p: it performs significantly better than the average for high p, and significantly worse than the MS measure for low p. But when (N, M) = (10, 50) (Fig. 4), the NC method is always better than the average with 99% statistical confidence, while it is not statistically different form the MS measure for p above 0.6.

Bottom Line: Our approach is based on the network relations induced between items by the rating activity of the users.Our method correlates better than the simple average with respect to the original rankings of the users, and besides, it is computationally more efficient than other methods proposed in the literature.Moreover, our method is able to discount the information that would be obtained adding to the system additional users with a systematically biased rating activity.

View Article: PubMed Central - PubMed

Affiliation: School of Business Administration, Northeastern University, Shenyang 110819, P.R.China.

ABSTRACT
We propose a new method for aggregating the information of multiple users rating multiple items. Our approach is based on the network relations induced between items by the rating activity of the users. Our method correlates better than the simple average with respect to the original rankings of the users, and besides, it is computationally more efficient than other methods proposed in the literature. Moreover, our method is able to discount the information that would be obtained adding to the system additional users with a systematically biased rating activity.

Show MeSH