Perbandingan Cosine Similarity dan Mean Squared Difference dalam Rekomendasi Buku Fiksi berbasis Item

Abstract

The need for recommendations is increasingly crucial in the digital era, especially with the abundance of fiction book data from e-book platforms and digital libraries. This study aims to evaluate the effectiveness of item-based collaborative filtering using cosine similarity and Mean Squared Difference (MSD) metrics for book recommendations. The knowledge discovery in databases method was applied, encompassing data selection, pre-processing, transformation, data mining, and evaluation. The dataset includes 100,000 user ratings obtained from Kaggle's "Book Recommendation Dataset." Our findings show that the Mean Absolute Error for MSD is 0.152307, slightly better than cosine similarity at 0.152406. The Root Mean Squared Error for MSD is lower at 0.185551, compared to cosine similarity's 0.185636. However, Cosine Similarity is more efficient in processing time, with 0.50 seconds compared to 0.59 seconds for MSD. Understanding these metrics is crucial, as they reveal differences in accuracy and efficiency in book recommendation. The results indicate that MSD performs better in the accuracy of fiction book recommendations compared to cosine similarity, making it more suitable for applications prioritizing recommendation precision, while Cosine is more efficient for large data processing.