Offline Evaluation Results of the Personalized News Recommendation System

The offline evaluation compared the personalized recommendation method with a latest-news baseline using five simulated user profiles and 100 news articles. The personalized method achieved higher Precision@5 (0.9600), nDCG@5 (0.8798), and Catalog Coverage (0.1800) than the baseline (0.4800, 0.3182, and 0.0500, respectively). However, its Category Diversity@5 was slightly lower (0.3600) than the baseline (0.4000). These results indicate improved recommendation relevance, ranking quality, and article coverage under the category-based relevance assessment, although recommendation diversity remains limited. The findings are restricted to the simulated profiles and pooled Top-5 recommendations.