Automatic Grading of Short Answers

ouahrani, Leila

Veuillez utiliser cette adresse pour citer ce document : http://dspace.univ-bouira.dz:8080/jspui/handle/123456789/17597

Affichage complet

Élément Dublin Core	Valeur	Langue
dc.contributor.author	ouahrani, Leila	-
dc.date.accessioned	2024-12-16T09:26:25Z	-
dc.date.available	2024-12-16T09:26:25Z	-
dc.date.issued	2024	-
dc.identifier.citation	Developing effective Automatic Short Answer Grading (ASAG) for e-learning environments is challenging. We consider scoring a text-constructed student answer compared to a teacherprovided reference answer. In this thesis, we address three key issues. The first key issue involves addressing the challenge of managing diverse student answers, considering that the reference answer may not cover their full range and often includes only specific correct responses. Secondly, developing an accurate grading model that enhances sentence similarity computation without requiring a large number of manually marked student responses is essential. Thirdly, it is crucial to ensure seamless integration into the Learning Management System (LMS) to enhance the practicality and scalability of the proposed system. The proposed solution addresses these challenges through three key components: a sequence-to-sequence deep learning network paraphrase generator, a supervised grading model, and an extension to the LMS quiz system. First, we provide a sequence-to-sequence deep learning network aimed at producing plausibly paraphrased alternative reference answers based on the provided reference answer. Second, we develop a supervised grading model that enhances features with specific and general course domain information using computational distributional semantics. Finally, we extend the question engine of the LMS quiz system to our model as a plugin for the open-source Moodle platform. Templates for creating and grading short answer questions have been successfully established and shared. Conducted in Arabic and English, quantitative experiments show that the paraphrase generator produces accurate paraphrases. The grading model yields comparable results to state-of-the-art and is deployed with low computational complexity to support short answers in online assessment. Two case studies were conducted in a real educational environment. The first case study resulted in the creation of the AR-ASAG dataset, which is the first publicly available Arabic dataset for ASAG evaluation. The second case study involved conducting a qualitative evaluation of a controlled class of students through formative and summative assessments using the proposed solution. The discussion covers the findings and implications, emphasizing valuable insights to advance the e-assessment of freetext short answers in online higher education.	en_US
dc.identifier.uri	http://dspace.univ-bouira.dz:8080/jspui/handle/123456789/17597	-
dc.language.iso	other	en_US
dc.publisher	Université de BOUIRA جامعة البويرة	en_US
dc.title	Automatic Grading of Short Answers	en_US
dc.type	Thesis	en_US
Collection(s) :	Faculté des Sciences et des Sciences Appliquées

Fichier(s) constituant ce document :

Fichier	Description	Taille	Format
Thesis Final OUAHRANI Leila - Progres.pdf		3,66 MB	Unknown	Voir/Ouvrir

Affichage abbrégé