Analisis Perbandingan Pearson Correlation dan Cosine Similarity pada Rekomendasi Musik berbasis Collaborative Filtering

Abstract

Advances in digital technology have revolutionized the world of music, making access to various genres and musicians easier and unlimited, but users still have difficulty finding music that suits their tastes. This research aims to analyze and compare the performance of the pearson correlation and cosine similarity methods on personal music recommendations based on Collaborative Filtering, with a focus on Item-Based Filtering, measured using Mean Absolute Error (MAE) and Root Mean Squared Error  (RMSE). The dataset utilized comprises public metal music ratings from Amazon, sourced from Kaggle, totaling 19,065 samples. The k-Nearest Neighbors (KNN) algorithm was employed for recommendation prediction. The research steps included data collection, pre-processing to address missing values, duplicates, normalization, and outlier detection, followed by prediction using the KNN algorithm, and accuracy measurement using MAE and RMSE. Evaluation results indicated that Pearson Correlation produced an MAE of 0.066538 and an RMSE of 0.086698, while cosine similarity yielded an MAE of 0.066559 and an RMSE of 0.086709. These findings suggest that pearson correlation is more effective in capturing linear relationships within the rating data, leading to recommendations that are more relevant and aligned with user preferences. Pearson correlation considers the variability in each user's ratings, resulting in more accurate recommendations that align with individual rating patterns.