A Novel Approach to Scalable and Automatic Topic-Controlled Question Generation in Education Conference: LAK25: The 15th International Learning Analytics and Knowledge Conference; March 03–07, 2025; Dublin, Ireland LAK25: The 15th International Learning Analytics and Knowledge Conference (LAK 2025),
Ziqing Li, Mutlu Cukurova, Sahan Bulathwela · 2025
Read A Novel Approach to Scalable and Automatic Topic-Controlled Question Generation in Education Conference: LAK25: The 15th International Learning Analytics and Knowledge Conference; March 03–07, 2025; Dublin, Ireland LAK25: The 15th International Learning Analytics and Knowledge Conference (LAK 2025), by Ziqing Li, Mutlu Cukurova, Sahan Bulathwela (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-05220.txt.
Introduction
Ziqing Li, Mutlu Cukurova, and Sahan Bulathwela published a 2025 paper on software that writes quiz questions on the topic a teacher actually wants. Automatic question generation could cut teachers' preparation load, but generic generators wander off-syllabus. The authors' method steers generation by topic, so a request about fractions yields fraction questions, at scale.
They fine-tuned a compact pre-trained language model on datasets built for education. Topic control is the added discipline: the model's output must match a specified subject, not just sound like a question. The paper was presented at a learning analytics conference, where the audience tests tools against classroom reality. Automatic generation will not replace teacher judgment, but it can draft the raw material teachers then select and polish. The authors frame the contribution as workload relief with relevance guaranteed. For large courses and question banks, that combination is the practical bottleneck this work attacks.
Catalog record
| CREATOR | Ziqing Li, Mutlu Cukurova, Sahan Bulathwela |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 50 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2501-05220
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A Novel Approach to Scalable and Automatic Topic-Controlled Question Generation in Education Conference: LAK25: The 15th International Learning Analytics and Knowledge Conference; March 03–07, 2025; Dublin, Ireland LAK25: The 15th International Learning Analytics and Knowledge Conference (LAK 2025), March 03–07, 2025, Dublin, Ireland DOI: 10.1145/3706468.3706487 ISBN: 979-8-4007-0701-8/25/03 CCS: Computing methodologies Natural language generation CCS: Computing methodologies Information extraction CCS: Applied computing Interactive learning environments Ziqing Li, Mutlu Cukurova, Sahan Bulathwela Abstract . The development of Automatic Question Generation (QG) models has the potential to significantly improve educational practices by reducing the teacher workload associated with creating educational content. This paper introduces a novel approach to educational question generation that controls the topical focus of questions. The proposed Topic-Controlled Question Generation (T-CQG) method enhances the relevance and effectiveness of the generated content for educational purposes. Our approach uses fine-tuning on a pre-trained T5-small model, employing specially created datasets tailored to educational needs. The research further explores the impacts of pre-training strategies, quantisation, and data augmentation on the model’s performance. We specifically address the challenge of generating semantically aligned questions with paragraph-level contexts, thereby improving the topic specificity of the generated questions. In addition, we introduce and explore novel evaluation methods to assess the topical relatedness of the generated questions. Our results, validated through rigorous offline and human-backed evaluations, demonstrate that the proposed models effectively generate high-quality, topic-focused questions. These models have the potential to reduce teacher workload and support personalised tutoring systems by serving as bespoke question generators. With its relatively small number of parameters, the proposals not only advance the capabilities of question generation models for handling specific educational topics but also offer a scalable solution that reduces infrastructure costs. This scalability makes them feasible for widespread use in education without reliance on proprietary large language models like ChatGPT. 1. Introduction One of the most significant and pertinent challenges facing education systems today is the teachers’ workload. It is argued to be the main reason behind issues associated with teachers’ retention in the profession as well as the lack of interest among graduate students to go into teaching professions. On the other hand, according to the Global Report on Teachers published by the Teacher Task Force and UNESCO, 44 million additional teachers will be needed by 2030 to meet Sustainable Development Goal 4 (SDG4), which aims to achieve universal primary and secondary education for all ( UNESCO and International Task Force on Teachers for Education 2030 2024 ) , without any improvements to the status quo . Among these concerns, generative AI in education is seen as an opportunity to ’transform a teacher’s day-to-day work’ ( for Education 2024 ) by reducing their workload and improving educational outcomes through the automation of routine tasks. Creating lesson materials and generating topic-specific, relevant, and age-appropriate questions for teaching have long been identified as time-intensive tasks for teachers, and an area where increased consistency is also expected to improve educational outcomes for students ( Giannakos et al. 2024 ) . Although learning analytics and AI in Education researchers have long explored ways to support teachers’ question generation capabilities through data-driven insights and models, attempts on Topic-Controlled Question Generation (T-CQG) have been less successful, primarily due to the lack of quality in the generated content. The use of large language models (LLMs) in teacher-facing interfaces, however, has the potential to address these quality concerns by leveraging recent advancements in NLP for automatic educational question generation (EdQG). EdQG can help teachers reduce the labor-intensive task of generating questions to promote classroom discussions, design formative and summative assessments, create lesson hooks, or address student misconceptions which are all activities that teachers consider among the most time-consuming in their profession ( for Education 2024 ) . Although most issues related to teachers’ workload are complex, ecosystem-level socio-technical challenges ( Cukurova et al. 2023 ) , T-CQG can serve as a small yet important practical step towards enhancing teachers’ productivity, aiming to mitigate workload and address issues related to teacher retention and attraction to the profession. In addition to their potential to support teachers, EdQG (and T-CQG) models can be integrated into learning management systems (LMSs) and intelligent tutoring systems (ITSs), to advance the system’s capability to perform precise diagnostics on learner’s knowledge gaps. The responses received from learners can inform the learning analytics pipeline more precisely and frequently to have a refined learner state representation, that can empower the system with targeted interventions. However, such interventions require advancements to generic neural network question generation models that do not have the ability to contextualise generation with constraints. The novelties that we introduce through this work are three-fold. We 1) propose a novel method to generate a dataset with contrastive examples in order to effectively train a T-CQG model and 2) validate and propose novel ways of evaluating the topical relatedness of the generations to the controlled topic using semantic relatedness metrics while 3) this is the only work that attempts in using a very small language model (sLM) with \approx 60M parameters, and succeeds in producing a T-CQG neural model. 2. Problem Definition, Background Research, and Research Questions In this section, we introduce the formal problem definition and prior work, leading to the research questions. 2.1. Problem Definition Although language models have been employed for question generation, their application in educational settings has only recently begun to be systematically explored with a heavy focus on the potential practical applications of proprietary models (e.g. GPT models’ prompt engineering and RAG applications for question generation). While existing research in relevant academic communities with a more technical focus explores generating questions from descriptive texts ( Du et al. 2017 ; Wang et al. 2018 ) , the task remains highly complex and there is less focus on the educational value of the generated questions in evaluations. Context plays a crucial role in the educational value of EdQG, yet much existing work has focused primarily on generating questions from sentences, paragraphs, or structured data in isolation ( Hu et al. 2018 ; Lopez et al. 2021 ) , with limited attention given to topic-controlled question generation in a given context. Topic-controlled question generation takes a target topic in addition to the descriptive text as context into account while generating the models’ outputs. On the other hand, traditional approaches which take sentences or paragraphs as inputs without contextual topic-control tend to generate questions that arbitrarily combine or select concepts and topics which are likely to be of limited practical value to professionals like teachers. From the learners’ points of view, prior research also suggests a strong correlation between the personalisation of testing and knowledge retention ( Bahrick et al. 1993 ) , which further supports the importance of topic-controlled question generation. Developing comprehensive, high quality and relevant educational question sets across different topics can significantly enhance teaching practice and support students through intelligent tutoring systems that provide personalised learning to diverse learners. In the scope of this work, we define topic-controlled question generation (T-CQG) as follows. Let us suppose a learner \ell has already consumed learning materials that contain the knowledge context c containing various topics T_{c} . A goal of a teacher or an intelligent system is then to generate a question