การเปรยบเทยบแบบจำลองการกรองขอมลสำหรบระบบแนะนำหลกสตร ระดบปรญญาตรภายใตขอจำกดความเบาบางของขอมล

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

This study aimed to compare the performance of data filtering models for an undergraduate program recommender system under data sparsity constraints. Seven recommendation models were compared and categorized into three main approaches: 1) Content-Based Filtering, consisting of Content-Based Filtering using the Probability-Based Matching approach; 2) Collaborative Filtering, consisting of User-Based Collaborative Filtering, Item-Based Collaborative Filtering, and Singular Value Decomposition (SVD); and 3) Hybrid Filtering, consisting of Hybrid (User-Based CF, CB), Hybrid (Item-Based CF, CB), and Hybrid (SVD, CB). The sample consisted of 417 high school students selected through purposive sampling. Participants were asked to choose their top five genuinely preferred undergraduate programs from a total of 49 programs. The ranked preferences were then transformed into scores to establish a 417 × 49 user–program matrix with approximately 89.8% data sparsity. The evaluation utilized the Top-N Recommendation approach at K= 1, 3, 5, and 10 using Hit Rate, Precision, Recall, F1-score, Mean Average Precision (MAP), and Mean Average Recall (MAR) as performance metrics. The results indicated that the Hybrid (SVD, CB) model with =0.80 achieved the best overall performance. In particular, at K=5, the model obtained Hit Rate@5 = 1.0000, Precision@5 = 0.8590, Recall@5 = 0.9153, F1@5 = 0.8863, MAP@5 = 0.9089, and MAR@5 = 0.5367. The findings suggest that integrating the SVD-based Model-Based Collaborative Filtering model with Content-Based Filtering can significantly improve the performance of undergraduate program recommender systems under sparse data constraints.