A Cost Adjusting Method for Increasing Customers’ Sentiment Classification Performance

Abstract

 Abstract—The internet could be a perfect platform for spreading the electronic word of mouth (e-WOM). Consumers not only heavily depended comments regarding the products or services in social media to make their purchase decisions. The negative product reviews could cause a negative impact on business products. When online reviews increase, inevitably there will produce imbalanced class data, in which the amount of positive comments (negative comments) is far larger than the number of negative comments (positive comments). When training a classifier using this kind of imbalanced data, it'll lead to a higher accuracy for determining the majority example, but an unacceptable error for classifying the minority examples. However, in the domain of sentiment classification, the available works didn't discuss this issue to solve the imbalanced comments. Therefore, this study aims to find the best combination from the cost adjustment, under-sampling, and over-sampling methods based on support vector machines (support vector machines, SVM) to improve the classification performance of imbalanced semantic comments. A comparative analysis of the experimental results will be provided to evaluation these methods. In addition, we use a real online travel site reviews as the case study to verify the effectiveness of the methods.