Improvement of Collaborative Filtering Recommendation System to Resolve Sparsity Problem using Combination of Clustering and Opinion Mining Methods
W.P. Krisdhamara, Bambang Pharmasetiawan, Kusprasapta Mutijarsa · 2019 International Seminar on Application for Technology of Information and Communication (iSemantic) · 2019
Online shopping or e-commerce sites are developing quite rapidly along with the development of the internet and technology in Indonesia. However, buyers in e-commerce still have problems trusting when buying products. Feedback from other buyers, either comments or ratings, is considered capable of influencing the behavior of buyers in buying items. In addition, a system of recommendations is known to help the buyer's decision to choose the product to be purchased. Collaborative Filtering (CF) is one of the popular recommendation methods. However, the ability of CF is limited by the problem of sparsity. In this study, we proposed model-based CF recommendation system that has good accuracy and quality recommendations by combining clustering methods using k-means++ to reduce data dimensions and opinion mining or sentiment analysis using Multinomial Naïve Bayes to filter recommendation results. K-means++ was chosen because it proved to have good quality clustering results, while Multinomial Naïve Bayes was chosen because of its simplicity and good accuracy in processing data used. This study used user-item interaction data and product reviews from Bukalapak. The experimental results showed that the proposed CF recommendation system had good accuracy with f-measure about 0.45 and quality improvement about 28%.