User-Based Collaborative Filtering Using
Agglomerative Clustering on Recommender System
Malim Muhammad
{malim.muhammad@gmail.com}
Department of Mathematics Education, Faculty of Teacher Training and
Education, Universitas Muhammadiyah Purwokerto
Abstract. Content-based, collaborative filtering, demographic, knowledge-based, and
hybrid recommender systems are the five categories of recommendation systems. User-
based collaborative filtering and item-based collaborative filtering are the two types of
collaborative filtering. However, the user-based approaches can be claimed to represent
the user; researchers will employ them here. This method is more concerned with the user's
likeness, or similarity than with the user's evaluated item. The accuracy of user-based
collaborative filtering approaches employing agglomerative Clustering with similarity
computations, i.e., cosine similarity, is improved in this study. MovieLens
(https://grouplens.org/datasets/movielens/) provided the researchers with the data they
needed. Between January 9, 1995, and October 16, 2016, a total of 100004 ratings for
9,125 films were collected from 671 individuals. At least 20 movies have been rated by
each user. Each rating has a value of 1 to 5. The data utilized for testing is five value data
from each user. In other words, 3,355 data points were tested in total. Using the single
linkage clustering approach to cluster films in the use-based method has been shown to
improve the accuracy of results that differ significantly between scenarios one and two,
namely 3,409 and 3.26. MAE and RMSE are the accuracy gauges utilized in the analysis,
and the smaller the value (closer to zero), the better the program results. The findings of
two trials (2 Scenarios) revealed significant differences between scenario 1 and scenario
2, namely 3,409 and 3.26. This is because in scenarios 1 and 2, only neighbors with
similarity values greater than zero are utilized to find predictions, regardless of whether
the neighbor has scored the film to be forecasted or not. In scenario 1, however, the results
produced by adding the single linkage clustering approach to the user-based method as
mentioned above are not as good. As the value obtained grows larger, the system's level of
accuracy decreases. However, the results achieved in scenario 2 are smaller, but the
differences are not significant.
Keywords: Recommended System, User-Based Collaborative Filtering, Agglomerative
Clustering
1 Introduction
Advances in technology have made the digital search easier. Over time various sites that
use search engines, be it selling sites or other sites, also use a recommendation system. The
recommendation system can be used in various areas, such as movies, news, music, books, and
others. According to Casey (2014), the recommendation system utilizes the history of user
behavior such as articles that have been read, products that have been assessed or purchased,
ISTED 2021, July 17-18, Purwokerto, Indonesia
Copyright © 2021 EAI
DOI 10.4108/eai.17-7-2021.2312410