Predicting repurchase intention using textual features of online customers reviews

Abstract

In an e-commerce setting, the large volume of online reviews may become a source of data to predict the repurchase intention. Repurchase intention is important for a company because it is related to customer loyalty. A machine-learning based methodology is proposed in this paper to perform the prediction of repurchase intention based on online customer reviews, in order to obtain the insights from a large volume of the available data. The preprocessing of the review texts, the tf-idf representation of words, and the feature selection using Fisher score are the essential components of the methodology. In the case study of an Indonesian cosmetic e-commerce website, it is shown that the inclusion of the textual content of customer reviews results in significantly higher prediction accuracies than using non-textual features (i.e, numerical ratings) only. Furthermore, for each category of repurchase intention labels (i.e., Yes, No, Maybe), a relatively distinct set of words that strongly predict the category is identified. To an extent, the sets reveal customers’ reasons that drive the stated repurchase intentions, which are useful for manufacturers to improve their products and design the marketing strategies.