I. Introduction
Automatic question classification (AQC) has been proposed for several applications and is achieved by defining domain-specific or domain-agnostic class labels. Domain-specific AQC is commonly applied in factoid question answering [1] with class labels being defined by subject matter (e.g., science, arts/humanities, business/finance) [2]–[4]. In contrast, domain-agnostic AQC is applied in information query or dialogue interactions in which the class labels may comprise question types (e.g., true/false, procedural, multi-part) [5] or reasoning capabilities (e.g., multi-hop, comparison, algebraic) [6], [7]. To enhance the effectiveness of deliberate practice [8], assessment questions are classified into their respective cognitive complexities (e.g., comprehension, synthesis, evaluation) for instructors to determine learners’ proficiency levels [9], [10].