<![CDATA[The growth of digital platforms offering a wide variety of content or products often leads to information overload, making it difficult for users to find items that match their preferences. This research aims to design and implement a content-based recommendation system capable of providing personalized recommendations based on the similarity of item characteristics. The methods employed include data pre-processing (case folding and text cleaning), text representation using Term Frequency–Inverse Document Frequency (TF-IDF), and the measurement of similarity between objects using Cosine Similarity. The dataset contains 133,102 book titles, with descriptive attributes converted into numerical vectors to form the basis of the recommendation process. The quality of the recommendations was evaluated using two complementary approaches: intrinsic metrics (average Cosine Similarity and Average Intra-List Similarity) and user-based validation via a questionnaire completed by 31 respondents who assessed 20 sample books, measured using Precision@5, Recall@5 and F1@5 under a leave-one-out protocol. The research results show that the system generates recommendations that are relevant to the reference objects and are confirmed by the preferences of real users. This approach is effective when applied in situations where there is limited user interaction data (cold-start).]]>
Copyrights © 2026
