In recent years online education has taken a flight: universities throughout the world have published their Massive Online Open Courses (MOOCs) on platforms such as Coursera and edX. While aforementioned platforms offer academics an online stage, other platforms can be described as marketplaces where basically anyone can become an online instructor. One example of such a platform is Udemy which offers more than 55,000 courses, just hit the 15-million students mark and is still growing rapidly.
This rise in popularity also comes with new challenges. Due to network effects and increased student demand platforms attract more and more creators which makes it for teachers even more challenging to differentiate from the many already existing courses. Creating a course from scratch requires a significant upfront time investment while the final pay-off remains uncertain. Teachers can therefore benefit from insights into what course characteristics are positively related to the average star rating and what makes a "best-seller" course. From a commercial standpoint platforms can also benefit from these insights. For example, to determine what courses to promote on their homepage. Further, Udemy has an interest in keeping students satisfied which becomes directly apparent from the many course reviews. By analysing and finding patterns in the abundance of reviews, they can support teachers in improving their current courses and developing even better ones in the future.
This project has been structured in such a way that both descriptive and predictive questions will be addressed. The overarching research theme can be translated into three main questions: to what extent can the course star review be predicted based on directly scrapeable course characteristics and teachers' prior activities on the teaching marketplace (1), what makes a best-seller course in the Data & Analytics subcategory (2) and to what extent can the sentiment of a course review (positive/negative) be predicted based on text features (3)?
The corresponding predictive models will be complemented with actionable descriptive insights into the platform dynamics which can support (aspiring) teachers to create high ranking, potentially best-seller, online courses.