At Televic Education we develop assessment and learning tools for universities, corporations, hospitals, government... For over a decade our users have been creating exercises and exam questions about a lot of different topics, such as healthcare, languages, math, biology... As a result, we now have a vast database of exercises about a lot of different topics.
It has always been the teacher/author who decides which set of questions the students or candidates have to solve, and when. But with this thesis we would like to investigate the possibility and usefulness of a system that automatically detects and annotates our exercises with a certain topic.
This would then allow students to select a topic of their interest (e.g. soccer, biology...) and automatically get random exercises about that topic.
Teachers or authors could also use these annotations to automatically construct exams or homework.
This data could also prove useful for other machine learning algorithms we want to use, such as automatic question generation, distractor generation and adaptive learning.
The goal of this thesis is three-fold:
1. research and compare the state-of-the-art algorithms on topic detection and decide on the best choice for our use case
2. develop a REST service that is able to categorize a question to a topic
3. evaluate the results on accuracy, precision and performance
Nature of the work
Specialty: AI/ Machine Learning, Software
Type of work: Research: 20% , Implem.: 60%, Experim.: 20%
Location: Televic, University
Type of activities: Experimenting, Implementation, Literature study, Measurements, Programming