Recommendation System Using User-Based Collaborative Filtering and Spectral Clustering

Recommendation systems are tools that can solve this problem by classifying people using User-Based Collaborative Filtering and Spectral Clustering approaches, resulting in more accurate recommendations. Preprocessing the data is the first step in the recommendation process, after which the data is grouped using the Spectral Clustering method. In the process of creating rating predictions and film recommendations based on similarity derived using the Pearson Correlation and Cosine Similarity algorithms, the clustering results are used to determine which users will be neighbors. Based on system experiments that have been conducted using variations in the number of clusters 3, 5, and 7, variations in the number of neighbors 1, 2, 3, 5, and 10, and comparing the results of MAE calculations of rating prediction results using a combination of spectral clustering methods with Pearson correlation, spectral Clustering with cosine similarity, with Pearson correlation and with cosine similarity, get results where the combination of methods with cosine similarity using 3 clusters and two neighbors becomes the method that has the best accuracy in making movie recommendations, namely with an MAE value of 0.3114. This is because the combination of methods has the smallest MAE calculation value. In other words, it has a minimal recommendation error rate. Meanwhile, the recommendation system with spectral Clustering only gets an MAE value of 0.3611.