The need to keep your skills up to date is becoming more and more essential in the current, permanently changing educational world. At the same time, the number of published educational content on the web is continuously increasing, while the lack of metadata and proper quality control of educational content is becoming an important issue for the search engine providers and educational recommender systems. This status quo is highly problematic for learners on the one hand when it comes to finding the most suitable educational material for their desired skills. On the other hand, this has also made the maintenance of learning pathways a frustrating job for curricula developers. In this research, we are proposing a novel Human-AI based recommender system, which combines a learning dashboard, and an open learning content/curriculum curation dashboard into one unified system to tackle the problem of individual learning path creation and maintenance both for curricula developers and learners.
In recent decades, we have faced a significant gap between the supply of learning content ofered
by educational systems, and what individuals actually need to learn to be able to carry out
their daily (including job-related and social-related) activities [
Subsequently, the regular updating of personal skill-sets and fast changes in the requirements
on the labor market side have made the process of creating and maintaining learning pathways
and individual curricula ineficient and time consuming for the learning content authors [
As a consequence, there is a need for the development of educational systems which connect learners with curricula developers eficiently. This system should 1. support curricula developers in creating up-to-date learning pathways, and 2. provide the fittest educational recommendations to the learners toward their learning goals.
In this research we plan to help curricula developers to create diferent pathways by providing insights on 1. skills required by the labour market, 2. learning topics that need to be covered to achieve a skill, and 3. high-quality educational content to cover learning topics. From the learners’ side, the created learning pathways are used to build a personalized learning environment for each individual learner. Therefore, the main objectives of this research are: • Proposing a method that facilitates the utilization of information on skills for learning processes, based on timely labor market information. • Decomposing those skills into meaningful learning objectives and their components (skills, learning topics) and ofering individualized learning pathways for learners. • Building a personalized educational content recommendation system that helps learners to achieve their goals through recommended high quality educational content. • Utilizing artificial intelligence to help authors and experts to create a high quality knowledge base, which serves as a foundation to define a vast number of learning pathways.
As a starting point to design our system, we created a three-level structure for the learners and content curators. Learning goals are on the top level, which consists of a set of skills. Skills are then divided into learning topics. In the end, educational contents are gathered and recommended based on learning topics for learners and curricula developers. This structure is shown in Figure 1.
Our approach to designing the aforementioned intelligent personalized recommender system with respect to our three-level structure consists of two modules: Content Curation and Learning.
The Content Curation module was built in four steps with respect to our three-level structure (i.e. learning goals, skills, and learning topics).
First, we created a content curation dashboard based on a three-level structure. Curators can add and maintain all three levels in the provided dashboard to create diferent paths for learners. These paths are dynamic and can change over time which helps the learners to be more up-to-date regarding their desired targets.
Second, we built a dynamic goal-skill matching component to monitor the changes in the
skill-sets required by the labour market. The component mines the text of online job vacancies
(as a proxy for learning goals), extracts the skill-related sentences and uses TFIDF to extract
the skills. Afterward, the extracted skills will be sorted based on their repetition in the past six
months to show their importance [
Third, we created a topic extractor method that decomposes a skill into learning topics.
We created models by applying various text-mining algorithms (e.g. TFIDF, Latent Dirichlet
Allocation, etc.) on the transcript of a large amount of educational content for a set of skills.
The component uses the created models to extract the topics from the educational resources
related to diferent skills [
Fourth, we developed an educational content management component to help them cover
the learning topics, which 1. collects educational content from diferent educational content
repositories, 2. classifies them according to their target subjects using our topic extractor
component, and 3. validates the quality of their metadata and content using [
The result of this AI-aided content curation module is shown in Figure 2.
First, we created a learning environment which is built on top of the learning goals, skills, and learning topics created and maintained by the curricula developers in content curation module (described in Section 2.1). Learners can choose learning goals as their target, study the required skills, and receive high-quality educational content for the respective learning topics. This environment also allows the learners to start learning a single skill or even a single learning topic.
Second, in order to personalize the educational content provided to each learner, we created a recommender system. This recommender system creates a user profile for learners based on some key user preferences (e.g., preferences regarding diferent educational formats, length of content, level of details, etc.). These preferences are gathered by asking questions from the learners during their registration in the system. The recommendation system will try to match the user profile with the metadata extracted from educational contents approved by content curricula developers. Moreover, learners provide feedback (rate their satisfaction) after completing each recommended educational content during the learning process. This helps our system to capture any changes and fine tune users’ profiles over time.
Based on our developed components, we built a prototype dashboard called eDoer1 which is available to everyone (a screenshot of a part of the learning environment is shown in Figure 3 and a screenshot of recommendations in the content curation environment is shown in Figure 4). In the curator dashboard of eDoer, curricula developers can define pathways by combining learning goals, skills, and learning topics. During the development and maintenance of these items, authors are aided with AI recommendations. Our system collects high quality educational content, which are relevant to the created learning topics. On the learning dashboard, each learner can search through all added curricula for their learning needs on diferent levels (learning goal, skill, or even learning topic). Then, they can add a list of learning goals, skills, and topics to their learning dashboard. In response, the system provides personalized educational content for each learner on the basis of their learning preferences. When content curators update curricula, all these updates are announced to the learners immediately so they can modify their learning pathway accordingly.
We have validated our learning dashboard in two diferent experiments. First, we did a
preliminary validation in the Business Analytics course at the University of Amsterdam. The
evaluation results showed that those students who used our prototype dashboard (24 out of
94 students used it voluntarily) achieved higher grades than those who did not use it [
1http://www.edoer.eu 2https://www.prolific.co This study is expected to empower lifelong learners to be able to autonomously work on their skill development. At the same time, it allows curriculum developers to define learning goals, skills, and learning topics faster, and keep up with emerging changes in the curriculum by getting help from artificial intelligence. Moreover, the automatic quality control component can help curriculum developers in validating the quality of their resources in various educational content repositories. By using our method, education providers are also able to collect open educational content with high-quality metadata and subsequently, ofer better recommendation and search services. These services help learners to spend less time and efort in finding related, high-quality educational content and also help authors to create and maintain the quality of their educational content more eficiently.
• ADSEE - Applied Data Science Educational Ecosystem, European Commission - Erasmus
Plus Programme • OSCAR - Online, open learning recommendations and mentoring towards Sustainable research CAReers, European Commission - Erasmus Plus Programme • BIPER - Business Informatics Programme Reengineering, European Commission - Erasmus Plus Programme • ADAPT - Implementation of an Adaptive Continuing Education Support System in the
Professional Field of Nursing German Federal Ministry of Education and Research • WBsmart - AI-based digital continuing education space for elderly care, German Federal
Ministry of Education and Research