IMPROVEMENT OF COLLABORATIVE FILTERING RECOMMENDATION SYSTEM TO RESOLVE SPARSITY PROBLEM USING COMBINATION OF CLUSTERING AND OPINION MINING METHODS

• Main Author: Widyas Pritha, Krisdhamara
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/39273
• Institution: Institut Teknologi Bandung

The rapid development of the internet in Indonesia is directly proportional to the development of the use of online buying and selling sites. Although e-commerce usage continues to increase, buyers still feel hesitant and afraid to buy goods or services through online sales sites. Therefore, every e-commerce site has implemented a review feature for products or sellers. Reviews are considered to affect customers to buy products rationally. In addition, the recommendation system is also stated to be able to assist buyers in deciding on purchasing products. The well-known recommendation system is Collaborative Filtering (CF). Although it has been widely implemented, CF still has problems with sparsity data. Research on the CF recommendation system on sparsity conditions has been carried out by applying clustering or opinion mining methods. This study aims to design a CF recommendation system that can provide accurate recommendations and have good quality too in sparsity conditions. This study uses model-based CF by applying clustering as a method to model data to reduce data dimensions. The k-means++ method for clustering was chosen because it was considered capable of producing better cluster quality. Then, opinion mining methods using Multinomial Naïve Bayes are applied with the aim of filtering recommendation items. The research experiment was conducted using review and transaction data from one of the e-commerce sites in Indonesia, namely Bukalapak. Evaluation of recommendations is measured using a f-measure and hit-rate. The evaluation results show that the CF recommendation system built has a good level of accuracy with a f-measure value of 0.45 and a quality increase of 28%.