Implementasi Content-Based Filtering Menggunakan TF-IDF dan Cosine Similarity untuk Rekomendasi Buku Akademik Mahasiswa

Abstrak

The abundance of digital book collections in academic libraries creates information overload, making it difficult for students to find relevant references. This study aims to design and evaluate an academic book recommendation system based on Content-Based Filtering (CBF) using TF-IDF for text feature weighting and Cosine Similarity for measuring inter-content similarity. The dataset consists of 10 simulated academic book entries from Kaggle, covering title, author, category, and description attributes across Computers, Technology Engineering, and Mathematics domains. Methodological stages include text preprocessing, TF-IDF feature extraction, similarity matrix construction, and top-N recommendation selection. Evaluation was conducted through subjective satisfaction testing involving 10 student respondents and quantitative evaluation using Precision@K and Recall@K metrics. Satisfaction results showed recommendation relevance (87%), system speed (90%), ease of use (85%), and overall satisfaction (86%). Quantitative evaluation revealed limitations with Mean Precision@5 of 0.0921, Mean Recall@5 of 0.0772, Mean Precision@10 of 0.0815, and Mean Recall@10 of 0.1136, exhibiting a consistent precision-recall trade-off. The system is concluded to be functionally effective but semantically limited. Future development is recommended to integrate word embedding techniques such as Word2Vec or BERT alongside hybrid filtering to substantially improve system performance.