BUDIJONO, SEBASTIAN AMADEUS DIAZ (2026) LIGHTWEIGHT STEREOTYPE-BASED COLD-START MITIGATION IN E-LEARNING VIA AHC AND DECISION TREE. S1 thesis, UNIVERSITAS KATOLIK SOEGIJAPRANATA SEMARANG.

Preview	Text 22.K1.0022 - SEBASTIAN AMADEUS DIAZ BUDIJONO-COVER_a.pdf Download (910kB) | Preview
	Text 22.K1.0022 - SEBASTIAN AMADEUS DIAZ BUDIJONO-ISI_a.pdf Restricted to Registered users only Download (1MB)
Preview	Text 22.K1.0022 - SEBASTIAN AMADEUS DIAZ BUDIJONO-DAPUS_a.pdf Download (919kB) | Preview
	Text 22.K1.0022 - SEBASTIAN AMADEUS DIAZ BUDIJONO-LAMP_a.pdf Restricted to Registered users only Download (11MB)

Abstract

E-learning platforms are new parts of education this decade, but their recommender systems often fail new users never interact with these apps previously, and the problem is called 'Cold Start'. This heavily affects the dropout rate of course enrollments, which leads to poor opinions. To address this critical gap, this study introduces a lightweight, stereotype-basedrecommendation system designed to deliver immediate, personalized content. Our approachbypasses the need for interaction history by leveraging explicit user attributes collected during sign-up, such as age, educational background, and learning goals. We employ a two-stagemethodology. First, we use Agglomerative Hierarchical Clustering (AHC) with Gower Distanceand Ward's method on a public MOOC dataset to identify and form distinct user "stereotypes"based on shared attributes. This process ensures that clusters, that are just made, are unique andvery different each other. Second, a Decision Tree classifier is trained on these stereotypes. When a new user joins, the decision tree, with its own trained rules, put the new user into good and fitting stereotype, then provide suitable courses inside the stereotype. This hybrid model offers anefficient, interpretable, and scalable solution to the cold-start challenge, significantly keeping thebest user experience without any heavier methods or algorithm.

Actions (login required)

View Item