MULTISTRATEGY APPROACH ON ACADEMIC RECOMMENDER SYSTEM

• Main Author: RACHMAWATI (NIM 23505044); Pembimbing: Ir. Dwi Hendratmo Widyantoro, M.Sc., PhD., EMA
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/18396
• Institution: Institut Teknologi Bandung

In academic domain, students are often having problem in choosing elective course during class registration. To solve this problem, students sometimes ask their friends or consult to their advisor about what optional subject should they take. Based on the situation, this thesis attempts to build an academic recommender system, a system that provides recommendation of elective course to students. The technique implemented at the system is a hybrid technique – a combination of collaborative filtering technique with content-based filtering technique – to produce a single recommendation. It was selected to take the advantage and eliminate any weakness of each technique. The third technique, a knowledge-based approach, was added in order to increase the recommendation quality. Item based algorithm was selected to avoid sparsity problem, a problem frequently occured in collaborative algorithm. Parsing and indexing process in content based was conducted by Lucene with its StandardAnalyzer. Then, the terms as the output of the process were weighted by using tf-idf (term frequency-inverse document frequency) method. The data used in the experiment was synthetic data. The elective course data was generated using assumption that students tended to take elective course which category was the same as that mandatory course having the highest mark. The experiment implements 12 hybrid strategies and 7 item based or content based strategies on 4 (four) synthetic data. The performance of each strategy was measured by using accuracy metric. The experiment on 4 (four) synthetic data showed that hybrid technic proved to be able to produce better accuracy than item based collaborative or content based technique. <br />