Content-Based Filtering Technique using Clustering Method for Music Recommender Systems

Abstract

The constant advancement of web development trends and technology has resulted in a large number of web systems that are frequently visited on a regular basis. Among the web systems that have been established include systems that allow users to listen to music online without having to download it to their devices. With the increasing popularity of music streaming, music recommender systems are important instruments for increasing digital music consumption. Machine learning (ML) is a form of artificial intelligence that makes the systems think like humans. ML allows a system to learn gradually to improve its accuracy in predicting future outcomes. The objective of this study is to develop a music recommendation system using one of the ML techniques, which is the content-based filtering technique. This study aims to explore on the music recommender system and how it is implemented, to design and develop a music recommender system. Popular algorithms for unsupervised learning, such as the k-means clustering, Euclidean distance, and cosine similarity methods were implemented in this study. These algorithms identify hidden patterns or data groupings without a human’s assistance. It is the best option for exploratory data analysis due to its ability to find informational similarities and differences. The system will determine the song feature values based on an analysis of the music user listens to during usage. This allows the algorithm to select similar songs after calculation in the database that would best match the user’s interests at any given time. K-means clustering was used to cluster the data according to the similarities of each song, separating them into different groups. Cosine similarity calculated the cosine distance with other data and recommended the one with a shorter distance. Euclidean distance calculated the direct distance between two vectors and recommended the one with a shorter distance. The results were then generated and presented to the user. Based on the findings, all the results produced by each method were accurate and similar.