The retrieved reading passage and a lexical dictionary are utilised to generate the remaining question components (Susanti et al. 2015).
According to Susanti et al. (2016), the word difficulty of the question components contributes to the difficulty of question items by up to 60%.
Unlike these past attempts, the automatic question generation system used in this work utilises Web texts from the Internet as well as information from WordNet to generate the question components (Susanti et al. 2015).
This study shows that item difficulty can indeed be controlled intrinsically by controlling the question components, thus suggesting the possibility of generating items automatically on-the-fly during the administration of a computerised adaptive test.
Their study revealed that (1) the word difficulty level of the question components contributed up to 60% of the item difficulty and (2) distractors had the greatest impact on item difficulty.
In this study, we propose a method that controls the item difficulty using three predetermined factors related to the question components: (1) reading passage difficulty (RPD), (2) semantic similarity between the correct answer and distractors (SIM), and (3) distractor word difficulty level (DWD).
Hence, the aim of the present study is to control item difficulty intrinsically, i.e. control the item difficulty through the characteristics of the question components (reading passage, correct answer, and distractor, as illustrated in Fig. 1) during the item generation process.
Thus, the question components of the items were the same for all three tests, but the response options for each item differed.
Nearly half of students (n = 70/160 or 43%) missed one or more question components that required finding an evidence-based citation.
According to the requirements of the project, such as if there are any signs of corrosion in the stretch of pipeline in question, components will be added to the smart pig before launch.
