**7. Related work**

Having in mind that our work in *OptSelectionAHP* is general enough to be applied in different domains, with performed evaluation analyses in domain of business process families, this section compares our work with work on (i) methods for representation and ranking of different kinds of requirements; and (ii) approaches for service and business process config‐ uration, including exact and heuristic techniques.

#### **7.1. Different kinds of requirements and prioritization algorithms**

Different researchers have been interested in developing tools and techniques for eliciting, formalizing and interpreting stakeholders' priorities over the existing options such as the pair‐ wise comparison method, priority groups, networks for decision‐making and cumulative ratings [3]. The selection of appropriate prioritization technique directly depends on its characteristics as well as the domain of application [32]: it is practical and universal to use qualitative concept to describe users' preference while the quantitative values corresponding to the qualitative concepts are not straightforward [33].

In our previous work [2], we gave the comprehensive review on methods for representation and prioritization of qualitative and quantitative preferences from different fields of research and standards. There, it is shown that one of the best‐known frameworks for addressing conditional preferences introduced by CP‐nets and TCP‐nets [3, 34] is not completely quanti‐ fied yet [35] and some other improvements need to be done in order to be used for effective ordering of decision outcomes.

On the other hand, there is a variety of methods based on quantitative measurements with supporting of only unconditional requirements, such as the Technique for Order Preference by Similarity to an Ideal Solution (TOPSIS) method [36], Simple Additive Weighting (SAW) [37], Linear Programming Techniques for Multidimensional Analysis of Preference (LINMAP) [38], Complex Proportional Assessment (CORPAS) [39], etc.

#### **7.3. Configuration of business processes and service selection approaches**

Many researchers have been studying the problems of business process configuration and service selection by using GA‐based solutions with different characteristics and elements.

The genetic approach for service selection and composition, proposed by Canfora et al. [22], uses one‐dimensional chromosomes and utilizes fitness function with penalty factor to select genomes and lead the convergence process to optimal solution. Similarly, GA‐based approach is proposed by Gao et al. [40], by using tree‐coded algorithm for service selection and compo‐ sition.

Furthermore, different studies and experiments are developed aimed on making comparisons of GA‐based solutions with other approaches widely used for solving optimal problems. Jaeger and Mühl [41] showed that GA‐based approach has better performance compared to Hill‐ Climbing (HC) approaches measured with both, reached overall quality and the closeness to the optimal solutions. Canfora et al. [22] conducted empirical research showing that GA‐based solutions are more scalable than Integer Linear Programming (ILP) solutions. Furthermore, they showed slower performance of GA‐based solutions which is significantly increased with larger number of available options.

However, the use of well‐known heuristic techniques for many NP‐hard problems (including ILP) have proven limitations [27] to be applied for the problem of service selection optimiza‐ tion, since various structural and semantic constraints cannot be handled straightforward. Additionally, the problem of business process families' configuration is more complex due to simultaneous selection of activities for which desired services should also be selected.

The application of genetic algorithms imposes additional concerns regarding specification of stakeholders' preferences regarding selection criteria aspects [22, 40]. Commonly used approach consider simple weighting schema (e.g., Simple Additive Weighting (SAW) [42]) for defining coefficients in the fitness function, which does not respect real‐world scenarios and the needs for complex weighting mechanism for the ranking and prioritizing stakeholders' requirements.
