Article Preview
TopIntroduction
In recent research areas of big data, cloud computing and IoT technologies, mining techniques should offer effective solutions to speed up entire applications. Web services perform an essential role in providing functionalities for exchanging information and services. Most electronic business (e-business) processes are endorsed by web services. The multiple web services are registered frequently in a central repository with descriptions.
In a typical scenario, a user’s request does not satisfy the single service or the perfect match of a requested service is unavailable in the discovery system. Multiple services needed to be composed and delivered to the user. When selecting a related service, the search engine finds multiple matches of the services. The discovery mechanism or agent therefore elects the best service among the available candidate services. The ranking involved in this situation produces the top-ranked results of the service description as a response to the user's request. Most users attempt the top-desired services on the list. The ranking process therefore is a substantial part of web service selection.
Furthermore, prediction is a crucial part of the web service selection. The system predicts the services that will help in the selection and business process. Many authors (Huang, Huang, Cheng, He, & Chen, 2017; Jayapriya, Mary, & Rajesh, 2016; Li, Wang, & Xiao, 2017) have intended to predict the service by developing algorithms in different areas of research. This is to improve the business by recommending services to the users. Users find their preferred services more easily because they are offered by the provider, which reduces the selection time. The curiosity of users also increases and motivates them to purchase other preferred services. These outcomes are derived from the experienced user’s perspectives or quality of service (QoS) values. The ranking of web service hence supports the efficient prediction system. Some questions raised in this stage include how to compare services, rank services, combine the entire rank list, determine the winner of the available service and produce a final list. Other questions raised include how to aggregate the top rank list given by experts, aggregate the full rank list given by all users and aggregate the partial rank list of missing services. This work picks up the challenge of predicting the rank sequence by aggregating available partial rank sequences that contain many services.
The ranking system generates a rank sequence with all participants of service. The collection of individuals’ input rank sequences (IRSs) are organized in two ways (Li et al., 2017), as shown in Table 1. Format 1 is item-based and each column represents the IRS of four individuals: IRS1, IRS2, IRS3 and IRS4. Each row represents the web services that partake in the ranking, whilst each cell represents the rank value of a particular web service given by individuals. Format 2 is rank-based and each column represents the IRSs of the four individuals mentioned above. Here, each row represents the rank value and each cell value represents the web services arranged in order as per the individuals. In the implementation, format 2 is more complicated for the aggregation of rank sequences. Format 1 thus is suitable for many real-world applications. The R package ‘RobustRankAggreg’ supports both format 1 and format 2. While giving input in format 2, first it is converted into format 1 and then it can be processed for aggregation. The ‘TopKList’ package processes the input in either of the above formats.
Table 1. The format of organizing the input rank sequence
Item-based | Rank-based |
Format1 | IRS1 | IRS2 | IRS3 | IRS4 | Format2 | IRS1 | IRS2 | IRS3 | IRS4 |
WS1 | 4 | 1 | 1 | 2 | 1 | WS3 | WS1 | WS1 | WS2 |
WS2 | 2 | 2 | 4 | 1 | 2 | WS2 | WS2 | WS4 | WS1 |
WS3 | 1 | 4 | 3 | 4 | 3 | WS4 | WS4 | WS3 | WS4 |
WS4 | 3 | 3 | 2 | 3 | 4 | WS1 | WS3 | WS2 | WS3 |