Hybrid Recommendation System Combining Collaborative Filtering and BERT-Based Semantic Embeddings

Views: 0
Downloads: 0
DOI https://doi.org/10.55923/aksara.v2i1.3
Nur Saputra 1 , Tri Saputra 2 , Dewi Kurniawan 3 , Wahyu Handayani 4 , Agus Saputra 5
1, 4, 5Department of Computer Science, Universitas Brawijaya , 2, 3Research Center for Informatics, National Research and Innovation Agency (BRIN)
1 nur.saputra@ub.ac.id , 2 tri.saputra@brin.go.id , 3 dewi.kurniawan@brin.go.id , 4 wahyu.handayani@ub.ac.id , 5 agus.saputra@ub.ac.id

Downloads

Cold-start challenges frequently impede standard collaborative filtering engines. By extracting deep semantic vector embeddings from item descriptions via BERT and fusing them with matrix factorization, our hybrid architecture enhances personalization accuracy by 14.3% in cold-start customer interactions.
Issue Vol. 2 No. 1 (2023)
Section Articles
Pages 89-101
Published 2023-06-30