A Comparative Analysis of Memory-based and Model-based Collaborative Filtering on the Implementation of Recommender System for E- commerce in Indonesia : A Case Study PT X P. H. Aditya, I. Budi, Q. Munajat 1 Faculty of Computer Science, University of Indonesia Depok, Indonesia pandhu.hutomo@ui.ac.id, indra@cs.ui.ac.id, qoribmunajat@cs.ui.ac.id Abstract The increasing growth of e-commerce industry in Indonesia motivates e-commerce sites to provide better services to its customer. One of the strategies to improves e-commerce services is by providing personal recommendation, which can be done using recommender systems. However, there is still lack of studies exploring the best technique to implement recommender systems for e-commerce in Indonesia. This study compares the performance of two implementation approaches of collaborative filtering, which are memory-based and model-based, using data sample of PT X e-commerce. The performance of each approach was evaluated using offline testing and user-based testing. The result of this study indicates that the model-based recommender system is better than memory-based recommender system in three aspects: a) the accuracy of recommendation, b) computation time, and c) the relevance of recommendation. For number of transaction less than 300,000 in database, respondents perceived that the computation time of memory-based recommender system is tolerable, even though the computational time is longer than model-based. Keywords e-commerce; collaborative filtering; recommender system; memory-based; model-based. I. INTRODUCTION It is important for e-commerce sites to provide innovative features to compete with others. There are three categories of e- commerce features which are Transactional, Relational, and Social [1]. Recommender system, as a relational feature, is one of important features that need to be implemented to improve the quality of e-commerce services. Some e-commerce sites that have implemented the recommenders are amazon.com and eBay [2]. Based on the method of implementation, recommender systems generally can be divided into two, memory-based and model-based. Memory-based method performs recommendation by accessing the database directly, while model-based method uses the transaction data to create a model that can generate recommendation [3]. By accessing directly to database, memory-based method is adaptive to data changes, but requires large computational time according to the data size. As for model-based method, it has a constant computing time regardless the size of the data but not adaptive to data changes. McCarey, Cinneide, and Kushmerick [4] conducted a study to evaluate memory-based and model-based collaborative filtering on software library. The research results show that memory-based approach is superior on two aspects which are precision and recall. Robillard and Walker [5] states that the nature of recommender system on software engineering and e- commerce domain are different. In the domain of software engineering, the recommendations are made by the task context, whereas in the domain of e-commerce, the recommendations are very dependent on the user's profile [5]. Considering the differences, it is necessary to study the implementation of recommender systems in e-commerce domain. Currently, there is still lack of studies that conduct comparative studies of model-based and memory-based recommender systems on the domain of e-commerce in Indonesia. Motivated by the growth of e-commerce industry in Indonesia, it is important for e-commerce sites in Indonesia to implement recommender systems to improve its service quality. However, there is still lack of studies that provide the best practices to implement recommender systems within the domain of e-commerce in Indonesia. Therefore, this study wants to give contribution by exploring two approaches of recommender system implementation which are memory-based and model-based collaborative filtering on e-commerce in Indonesia. In order to perform the study, one e-commerce company in Indonesia is selected as a case study. The performance of each method is evaluated based on the computation time, accuracy, and relevance of the recommendation. To explain the conduct of the study, the paper is structured as follows. We first explained the context of e-commerce and the related theories in recommender systems followed by the research methodology explanation. Then, we explained the implementation of the recommender systems followed by the evaluation. The results and analysis is discussed in section 5 and finally section 6 conclude the finding of this research. II. RECOMMENDER SYSTTEMS IN E-COMMERCE This study applied recommender systems within the domain of e-commerce. Kalakota and Whinson [6] defines e-