Vocational choice tests, in their current state, often fall short in delivering personalized and engaging recommendations, resulting in a lack of support during crucial choice moments at school. As a result, students rely heavily on parental advice, potentially limiting later career satisfaction. This research project aims to harness advanced machine learning techniques, such as (deep) reinforcement learning (RL), to offer tailored and improved vocational recommendations considering the individual student’s interests.
Recommender systems can be used to match specific vocations or career options to the user, based on the user profile and the characteristics of the vocational choice options. Leveraging the ability of deep learning (DL), deep reinforcement learning (DRL) can be applied in large action spaces, such as vocational choice tests. On the downside, because they use DL, deep RL algorithms and systems are less explainable than conventional RL systems, thereby potentially creating a black-box system.
The primary objective of this PhD project is to design and develop recommender systems that employ innovative algorithms to deliver personalized vocational guidance. By incorporating sophisticated machine learning, we aspire to create a user-centric approach that considers individual interests and preferences. Resulting matches have to be explainable to the user. Therefore, in this PhD project the black-box of DRL algorithms will be opened, for example through simultaneous explanation and policy learning.
Moreover, we aim to integrate feedback mechanisms that allow users to shape and refine their recommendations further. This innovative approach not only enhances the usability of vocational tests but also empowers individuals to make better career choices.
The main research goals are:
- Designing and implementing explainable (deep) reinforcement learning algorithms for recommender systems.
- Evaluating algorithm performance through (1) user opinions on the recommendations, (2) user opinions on the vocational choice test itself, and (3) long term satisfaction and relevance of recommendations (validity).