S3-CT4-2 - OER Recommendations to Support Career Development3. Research Work In Progress
1 German National Library of Science and Technology (TIB)
2 RWTH Aachen University
3 University of Amsterdam (UvA)
1 Research Problem
This Research Work-in-Progress Paper departs from the turbulent changes on global labour markets, which force learners to think autonomously about their individual skill targets. Learners therefore 1) need to be equipped with skills to be autonomous, and strategic about their own skill development, and 2) they need high-quality, personalized educational content and services.
Open Educational Resources (OERs) can contribute to the resolution of these problems as they are available in a wide range of contexts globally. However, the applicability has been limited, due to low metadata quality and complex quality control. These resulted in lack of personalised OER recommendation and search functions.
Therefore, we suggest a novel, personalised OER recommendation method to match skill demands with open training content through:
- Exploratory data analysis to enhance the quality OER metadata
- Creating an OERs quality prediction model
- Aiding learners to set individual skill targets based on labour market information
- Building a personalized OER recommender to help learners to master their target skills.
By applying Natural Language Processing techniques, we constructed a job - skill matching model based on vacancy announcements. Subsequently, we crawled 9,728 OERs from different sources to analyse their metadata, and established a relationship between OER metadata and content quality. We built an OER quality prediction model based on 6 different properties (source, quality probability, popularity, skill similarity, length, accessibility).
For effective recommendations, our system uses similar 6-dimensional preference-vector for learners together with their job and skill targets, and contextual information (location, discipline) to initialize learning preferences. OER properties are initialized similarly, by using and benchmarking a number of metadata (e.g. subjects, authors).
Based on the above mentioned properties, OERs are recommended to learners. Learners interact with the recommender through a dashboard, in which they can search for their desired job, display the list of required skills, set their level of expertise for each skill, and access relevant OERs. During their learning process, learners rate their satisfaction with recommendations, and update their learning preference-vector. This strategy detects changes in learner profiles and fine tune the precision of recommendations.
We evaluated our prototype with 23 in-depth, semi-structured interviews with experts to learn how our prototype performs in terms of its objectives, logic, and contribution to learning. Experts generated more than 400 recommendations. 80.2% of these recommendations were reported as useful. Interviews revealed that our recommender shows potential to improve the learning experience of lifelong learners.
This study is expected to 1) empower learners to take control and responsibility for their own skill development, 2) improve skills on the basis of labour market information and personalised OER recommendation, and 3) progress the literature on OER quality control. We expect that learners will show enhanced self-regulation, thus spend less effort to find relevant, high-quality OERs. This approach also contributes to OER property identification accuracy, which is essential to increase OER (re)usability.