|Title||:||Generating Valid Multiple-Choice Questions from SHIQ ontologies by considering role-restrictions and open-world assumption.|
|Speaker||:||Vinu E V (IITM)|
|Details||:||Tue, 2 Aug, 2016 2:30 PM @ BSB 361|
|Abstract:||:||Knowledge with potential educational value is now widely available in the form of Description Logics (DL) based Ontologies. However, using this new knowledge in e-Learning systems is an advancing area of research. One of the challenging functionality of an unsupervised e-Learning system is to conduct assessment test by automating the creation of specific type of assessment items.
In our research, we explore the use of SHIQ DL ontologies in generating question items. In this talk, the prime focus will be given to generating multiple-choice questions (MCQs) from ontologies which have rich terminological axioms (for example, ontologies like NCI, Plant-Protection etc. where there are rich concept definition & inclusion axioms). There are no existing works in the literature which fully utilize the potential of terminological axioms (a.k.a Tbox) in question generation (QG). We observed that, in the state-of-the-art techniques the role restrictions present in the Tbox axioms were ignored while framing questions, leaving them with questions which are only based on concept inclusion. Our studies show that inclusion of semantically-refined role restrictions is the key to generating question statements that are close to human generated ones.
Further, we will focus on how to generate valid distractors for generated question statements. Distractors which are selected randomly cannot be guaranteed to be the wrong answers of a given stem. In such cases, human experts are required for validating the generated question items. We propose an approach, which works with open-world assumption, to generate valid choice-sets, so that an unsupervised e-Learning systems can be built. The method ensures that choice-sets consist of distractors that are semantically close to the correct answer. We have implemented a prototype system, and studied the feasibility of the proposed approaches. The details of an empirical study and other promising applications of the Tbox-based QG process will be also presented.