Table 1.
Online cloud storage services.
Table 2.
A simple customer-service matrix.
Table 3.
Significances arising from the experience usability.
Table 4.
Significances arising from the value distribution.
Table 5.
Similarities between different services.
Table 6.
Predicted attribute values.
Figure 1.
Customer satisfaction function Cm,n.
(a) and (b) depict the distributions of customer satisfaction as recorded at the fixed expectation Hexp = 0.7 and Hexp = 0.9, where the parameter δ is varied from 2 to 6 in increment of 2. It can be observed that the rate of change in customer satisfaction differs significantly when Hm,n falls below and exceeds the expectation.
Table 7.
Customer satisfaction.
Figure 2.
(a) and (b) depict the value distributions of response-time and throughput in our customer-service matrices, where “−1” indicates that the service invocation failed due to an http error. The ranges of response-time and throughput are 0–16.053 seconds and 0–541.546 kbps, respectively.
Figure 3.
The number of recommended services for u339.
Results are presented for the proposed cloud service selection approach, where the parameter is varied from 0 to 1 in increment of 0.1.
Table 8.
Trustworthy attributes of cloud service.
Figure 4.
Impact of preference and expectation.
(a) and (b) depict the experimental results of preference parameter δ and expectation Hexp, respectively. They indicates that δ regulates the elimination rate of untrustworthy cloud services, whereas Hexp controls the degree of customer's tolerance to untrustworthy service.
Figure 5.
Impact of neighborhood size k.
(a) and (b) depict the MAE fractions of JV-PCC, f-PCC and IPCC for response-time and throughput, while (c) and (d) depict the RMSE fractions. It can be observed that JV-PCC achieves smaller MAE and RMSE consistently than f-PCC for both response-time and throughput. Regardless of JV-PCC or f-PCC, as k increases, MAE and RMSE drop at first, indicating that better performance can be achieved by employing more similar services' records to generate the predictions. However, when k surpasses a specific level (i.e. k = 25), they fail to drop with a further increase in k, which were caused by the limited number of similar neighbors.