Hybrid Book Recommendation System  Using Content-Based Filtering and Collaborative Filtering Based on Singular Value Decomposition

• Atika Pratiwi Harahap

  Malikussaleh University

  Author

Introduction: The rapid growth in the number of digital books on platforms such as Goodreads, Google Books, and Amazon has led to information overload and a paradox of choice for readers. The book recommendation system is an important solution to provide personalized and relevant advice. 
Objective: This study aims to develop a hybrid book recommendation system using content-based filtering and collaborative filtering based on singular value decomposition.
Methods: This study developed a hybrid book recommendation system that combines TF-IDF-based Content-Based Filtering with Collaborative Filtering based on Singular Value Decomposition (SVD). The Goodbooks-10k dataset (10,000 books, 981,756 ratings from 53,424 unique users) was used in this study. In Content-Based Filtering, text features are extracted from a combination of tags, titles, and authors using the TF-IDF Vectorizer (max_features = 5,000, ngram_range = (1,2)) and similarity is calculated by cosine similarity. Collaborative Filtering uses SVD with 50 latent factors on the normalized user-item matrix (14,639 × 9,999). 
Results: The results of the evaluation showed that Content-Based Filtering had a diversity of 0.7250 but low coverage (0.0029) due to popularity bias, while SVD-based Collaborative Filtering obtained an RMSE of 3.5613 and MAE of 3.3896 in 1,000 random test samples.
Conclusion: The hybrid system developed can overcome the limitations of each single method to produce more accurate, personalized, and diverse recommendations. This research contributes to the development of a computationally efficient digital literature recommendation system.

• pdf

• 2026-01-02 (5)
• 2025-12-31 (4)
• 2025-12-31 (3)
• 2026-01-02 (2)
• 2025-12-31 (1)