Multiple-objective optimization applied in extracting multiple-choice tests. (October 2021)

• Record Type: Journal Article
• Authors: Nguyen, Tram; Bui, Toan; Fujita, Hamido; Hong, Tzung-Pei; Loc, Ho Dac; Snasel, Vaclav; Vo, Bay
• Is Part Of: Engineering applications of artificial intelligence. Volume 105(2021)
• Publication Date: 2021-10
• Subjects: Multiple-choice test -- Test construction -- Multiple objective optimization -- Test-question bank -- Simulated annealing -- Genetic algorithm; Engineering -- Data processing -- Periodicals; Artificial intelligence -- Periodicals; Expert systems (Computer science) -- Periodicals; Ingénierie -- Informatique -- Périodiques; Intelligence artificielle -- Périodiques; Systèmes experts (Informatique) -- Périodiques; Artificial intelligence; Engineering -- Data processing; Expert systems (Computer science); Periodicals; 620.00285
• DOI: 10.1016/j.engappai.2021.104439
• Languages: English
• ISSNs: 0952-1976

Abstract

Abstract: Student evaluation is an essential part of education and is usually done through examinations. These examinations generally use tests consisting of several questions as crucial factors to determine the quality of the students. Test-making can be thought of as a multi-constraint optimization problem. However, the test-making process that is done by either manually or randomly picking questions from question banks still consumes much time and effort. Besides, the quality of the tests generated is usually not good enough. The tests may not entirely satisfy the given multiple constraints such as required test durations, number of questions, and question difficulties. In this paper, we propose parallel strategies, in which parallel migration is based on Pareto optimums, and applyan improved genetic algorithm called a genetic algorithm combined with simulated annealing, GASA, which improves diversity and accuracy of the individuals by encoding schemes and a new mutation operator of GA to handle the multiple objectives while generating multiple choice-tests from a large question bank. The proposed algorithms can use the ability to exploit historical information structure in the discovered tests, and use this to construct desired tests later. Experimental results show that the proposed approaches are efficient and effective in generating valuable tests that satisfy specified requirements. In addition, the results, when compared with those from traditional genetic algorithms, are improved in several criteria including execution time, search speed, accuracy, solution diversity, and algorithm stability.