Towards Data-Driven Sequencers in Intelligent Tutoring Systems

In Intelligent Tutoring Systems (ITS), Artificial Intelligence and Machine Learning are increasingly used as decision-making methods. Performance Prediction is one of their main subtasks whose final goal is to predict student performances and define the students’ competence in relation to specific skills. Adaptive sequencers take a student’s past performances into account in order to select the next task which best fits the student’s learning needs.

In all the aforementioned subtopics, iTalk2Learn is working to ameliorate state-of-the-art. The University of Hildesheim’s (UHi) work, starting from previous contribution to Performance Prediction Methods [3-6], is focused on sequencing methods [1].

Simple task scheduling is based on fixed sequences decided by a human expert. Adaptive policies instead rely on assumptions such as that a student will be able to successfully complete the exercises of the achieved difficulty level but not the more difficult ones without having completed the ones of the previous level. Empirical observation suggests that this can be problematic as it requires students to go through all the topics in the current level even if they can answer them successfully in their first attempt. Although the power-law-of-practice would suggest that students should be provided with several opportunities to practice, unnecessary repetition can be detrimental in that it can lead to frustration and influence the student’s perception of the reliability of the system.

One approach to the problem is based on assessing the students’ skills and matching them to the required skills and difficulties of the available tasks. For example, the skills that are not as well developed are selected to be practiced in the next session. In this scenario two problems arise:

  1. Tagging tasks with required skills necessitates experts and thus is time-consuming, costly, and, especially for fine-grained skill levels, also potentially subjective.
  2. Learning adaptive sequencing models requires online experiments with real students and specific data collection policies that consist, at the beginning, in many randomly proposed tasks.

Our main goal is to take the state-of-the-art in sequencing educational contents further, while achieving a solution that relies on data rather than expertise and does not place too many demands on data collection modalities and students’ effort. As a result, this work allows easier integration of a data-driven sequencer into an already existing ITS, supporting our current work at the iTalk2Learn project, which aims to incorporate content from different systems into an open architecture [2].

In the paper, ‘Adaptive Content Sequencing without Domain Information’ [1], we showed how a score prediction method and a simple policy, inspired by Vygotsky’s concept of Proximal Development, could be used to ameliorate sequencing in a simulated environment.

In order to do so we developed:

  1. A content sequencer, the Vygotsky Sequencer, based on a performance prediction system that (1) can be set up and firstly evaluated in a laboratory, (2) models multiple skills and individualisation without engineering/authoring effort, and (3) adapts to the combination of contents, levels and skills available.
  2. A simulated environment with multiple skill contents and students’ knowledge representation, where knowledge and performance are modeled in a continuous way.

We are currently working on integrating the Vygotksy Sequencer into the Whizz platform, one of iTalk2Learn use cases, to run an experiment with one hundred 8-year-old children in London sequencing lessons of three different topics. The integration between the Vygotsky sequencer and the Whizz platform will exploit the novel sequencing-framework presented in the paper ‘Minimal Invasive Integration of Learning Analytics Services in Intelligent Tutoring Systems’ [2].


References:

[1] Schatten C. and Schmidt-Thieme, L. (2014) Adaptive content sequencing without domain information. In proceedings of CSEDU 2014.

[2] Schatten C., Wistuba M., Schmidt-Thieme, L. and Gutiérrez-Santos S. (2014) Minimal Invasive Integration of Learning Analytics Services in Intelligent Tutoring Systems, 14th IEEE International Conference on Advanced Learning Technologies – ICALT2014.

[3] Thai-Nghe, N., Drumond, L., Krohn-Grimberghe, A., Schmidt-Thieme, L. (2010) Recommender system for predicting student performance. In Procedia Computer Science, 1(2), 2811-2819.

[4] Thai-Nghe, N., Drumond, L., Horvath, T., Krohn-Grimberghe, A., Nanopoulos, A., Schmidt-Thieme, L. (2011) Factorization techniques for predicting student performance. In Educational Recommender Systems and Technologies: Practices and Challenges. IGI Global (2011).

[5] Thai-Nghe, N., Drumond, L., Horvath, T., Nanopoulos, A., Schmidt-Thieme, L. (20100) Matrix and Tensor Factorization for Predicting Student Performance. In CSEDU (1).

[6] Thai-Nghe, N., Horváth, T., Schmidt-Thieme, L. (2011) Personalized forecasting student performance. In Advanced Learning Technologies (ICALT), 2011 11th IEEE International Conference on. IEEE.


 

  • Jimmie Cordeiro

    For me, this will be such a great idea that they have able to promote because through this kind of research, it will just easy for them to determine the ability of a certain students as well as their intelligence.