<![CDATA[The rapid development of digital libraries and online bookstores has increased the need for intelligent book recommendation systems that can understand user preferences and provide relevant suggestions. However, many existing systems rely on simple keyword matching or collaborative filtering, which often fail to capture the semantic meaning of complex user descriptions. This study aims to develop and evaluate a content-based book recommendation system that combines Word2Vec word embedding models with Knowledge-Based Filtering to improve the relevance of recommendations based on user-provided descriptions. The proposed system utilizes two Word2Vec architectures, Continuous Bag of Words (CBOW) and Skip-gram, to learn semantic relationships between words in book descriptions and user inputs, while Knowledge-Based Filtering incorporates explicit attributes such as publication year, genre, author, and book length to refine the results. The system was tested using a descriptive query: “a good fiction story telling a boy school at great magic school, published on 1995-1999”. The evaluation, measured by Precision@K and Recall@K at K = 5, 10, and 20, shows that CBOW outperformed Skip-gram, achieving a perfect Precision@5 of 1.00 and balanced precision and recall at higher K values, while Skip-gram exhibited more variability at small K. These results indicate that CBOW is more effective in providing stable and highly relevant recommendations at the top of the list. This research confirms that combining semantic embedding and knowledge-based approaches enhances the accuracy and flexibility of recommendation systems. Further studies can explore diverse datasets and user interfaces to broaden practical applications in digital library and e-commerce platforms.]]>
