CDGRS-LOD model: leveraging linked open data in collaborative cross-domain group recommender system

This study introduces a novel Cross-Domain Group Recommender System leveraging Linked Open Data (LOD) (CDGRS-LOD), designed to improve the accuracy and relevance of group recommendations. Group recommender systems (GRS) are increasingly used to deliver collective recommendations, but their effectiveness is often constrained by sparse user data and limited semantic context during group formation. The proposed CDGRS-LOD model enriches movie metadata with structured external knowledge, incorporating cross-domain attributes such as the Music Composer feature extracted from DBpedia via SPARQL queries. This enhancement improves the model’s understanding of relationships between users and items, thereby enabling more accurate and contextually relevant group recommendations. Singular Value Decomposition (SVD) predicts missing ratings from the enriched metadata, effectively mitigating sparsity. Two sparsity reduction levels, 1% and 5%, were evaluated, with the 5% reduction yielding the best performance and subsequently used for group formation with a k-Nearest Neighbours (kNN) model. The model is evaluated against existing methods, including GRS-LOD, Archi2_bkm, Archi2_km, and Predict & Cluster. Results show that CDGRS-LOD achieves a Root Mean Square Error (RMSE) of 0.5917, outperforming GRS-LOD (0.7816) and traditional approaches (0.9435–0.9781), representing up to 24% improvement over the best baseline. These findings confirm that integrating cross-domain knowledge with LOD enhances metadata richness, reduces sparsity, lowers prediction errors, and improves group recommendation quality.