E-commerce Recommender System Using PCA and K-Means Clustering

Dendy Andra,Abdurahman Baizal Baizal

doi:10.29207/resti.v6i1.3782

Dendy Andra, Abdurahman Baizal Baizal

Open Access

https://doi.org/10.29207/resti.v6i1.3782

Copy DOI

Abstract

Recently, recommender system has an important role in e-commerce to market products for users. One of recommender system approach that used in e-commerce is Collaborative Filtering. This system works by providing product recommendations based on products liked by other users who have similar preferences. However, sparse conditions in user data will cause sparsity problems, namely the system is difficult to provide recommendations because of the lack of important information needed. Therefore, we propose an e-commerce product recommendation system based on Collaborative Filtering using Principal Component Analysis (PCA) and K-Means Clustering. K-Means is used to overcome sparsity problems and to form user clusters to reduce the amount of data that needs to be processed. While PCA is used to reduce data dimensions and improve clustering performance of K-Means. The test results using the sports product dataset on the Olist e-commerce show that the proposed system has a lower RMSE value compared to other methods. For the number of neighbors of 10, 20, 30, and 40, our system obtains values of 0.771806, 0.75747, 0.75304, 0.75304, and 0.75270.

Highlights

Recommender system is an important part of a business strategy in e-commerce [1]
In this study, we propose a recommender system for e-commerce products based on Collaborative filtering using Principal Component Analysis (PCA) and K-Means Clustering methods
We tested the effect of the number of PCA components on the clustering performance of K-Means to ensure the optimum number of components

Summary

Introduction

Recommender system is an important part of a business strategy in e-commerce [1]. Recommender system is used to provide product recommendations to users with the aim of helping users get the desired product [2]. Collaborative filtering (CF) is a recommendation system method used in e-commerce. This CFbased system generates recommendations by utilizing information from other users who have similar preferences. The incompleteness of user-related data, such as a lot of empty or unavailable data creates a sparsity problem, that is the lack of important information for recommender system to produce recommendations for users [3]. The accuracy of recommendations will decrease and the system cannot generate recommendations that are relevant to the user. This inaccuracy of product recommendations to users can lead to decreased user confidence in the system [4]

Methods

Results

Conclusion