Evidence record · UDINUS · SDGs
- Unit
- Rektorat UDINUS
- Period
- 2023
- Goals
- Record
- fad1da4c-c5e4-471f-9b6b-dfd45f53dbc0
- Published
- 2026-10-30
Universitas Dian Nuswantoro conducted a research activity on adopting machine learning in demographic filtering for a movie recommendation system. The activity involved Yin, Lee Jia; Safar, Noor Zuraidin Mohd; Kamaludin, Hazalila; Abdullah, Noryusliza; Yusof, Mohd Azahari Mohd; and Supriyanto, Catur as participants. The activity addressed two issues in machine learning movie recommendation systems: cold start and data sparsity. It sought a decision-making algorithm for addressing the complex start problem through precise parameters.
The activity implemented a demographic filtering technique with the k-means clustering method. The findings examined the effects of demographic filtering on movie recommendations. Demographic filtering grouped users into clusters based on gender, age group, and occupation. The cluster distribution represented groups based on the top 100 results of the experiment, and the user with the least distance to the cluster centre was selected as the usual group representative in each cluster.
Three clusters were experimented with. Cluster 0 represented males who were college or graduate students aged 25 to 34. Cluster 1 represented females in executive or managerial occupations aged 25 to 34. Cluster 2 represented males in sales or marketing occupations aged 35 to 44. The activity also found that users from different clusters had different preferred movie genres. Cluster 0 preferred action, adventure, comedy, drama, and war. Cluster 1 preferred comedy, crime, drama, horror, romance, and sci-fi. Cluster 2 preferred action, comedy, drama, film-noir, mystery, and thriller.
The activity contributed to demographic filtering studies as an alternative solution for future technical development work. This institutional research activity documents the implementation of demographic grouping and k-means clustering in a movie recommendation system, together with the reported cluster characteristics and genre preferences.