Component-based thinking in designing interactive multiobjective evolutionary methods

Multiobjective optimization problems have multiple conflicting objective functions to be optimized simultaneously. They have many Pareto optimal solutions representing different trade-offs, and a decision-maker needs to find the most preferred one. Although most multiobjective evolutionary algorithms approximate the Pareto optimal set, their variants incorporate preference information to focus on a subset of solutions that interest the decision-maker. Interactive methods allow decision-makers to provide preference information iteratively during the solution process, enabling them to learn about available solutions and their preferences' feasibility. Nevertheless, most interactive evolutionary methods do not sufficiently support the decision-maker in finding the most preferred solution and may be cognitively too demanding. We propose a framework for designing and implementing interactive evolutionary methods. It contains algorithmic components based on similarities in the structure of existing preference-based evolutionary algorithms and decision-makers' needs during interaction. The components can be combined in different ways to create new interactive methods or to instantiate the existing ones. We show an example of the implementation of the proposed framework composed of three elements: a graphical user interface, a database, and a set of algorithmic components. The resulting software can be utilized to develop new methods and increase their usability in real-world applications.