- 期刊
- OpenAccess
Feature selection using binary particle swarm optimization algorithm to predict repurchase intention from customer reviews
摘要
Indonesia has the most prominent digital economy in Southeast Asia and has a promising market for e-commerce companies to compete and dominate the online market share. This also gave rise to an increment in the number of customer reviews of a product or service provided. Online customer reviews can be utilized to analyze the repurchase intention of e-commerce customers. However, many features appearing in customer reviews increased the repurchase intention predictive model complexity. A process to choose a subset of features and reduces the number of features in data is called feature selection. This paper proposed a method of feature selection to pre-process the inputs to the predictive model. The selection is performed using a metaheuristic called Binary Particle Swarm Optimization (BPSO) combined with Sentiment Orientation-Pointwise Mutual Information to sort the features. The sorting corresponds to the particle dimension, which is a part of the particle encodings that affect the metaheuristic's performance in solving the problem. The results show that the proposed method reduces and selects the best features to construct a predictive model of repurchase intention from online customer reviews on two datasets that are written both in Indonesian and English. Compared to the baseline model before performing feature selection, the accuracy of the predictive models evaluated using k-Nearest Neighbors on both datasets increased by 5.40% (75.91% to 81.31%) and 8.50% (71.37% to 79.87%), respectively.