A novel fuzzy nonparallel support vector machine for identifying helpful online reviews

Online review datasets are always imbalanced and contain numerous outliers or noise, making the accurate and efficient identification of helpful reviews a critical challenge in the digital age. To address this issue, the optimal feature set is first obtained from numerous constructed possible features (including ones based on the knowledge adoption model) by a feature selection method, and then a novel fuzzy nonparallel quadratic surface support vector machine (FNQSSVM) model is proposed for identifying helpful online reviews in this study. For well handling the imbalanced data with outliers or noise, a novel fuzzy membership function is first developed based on the K-nearest neighbor method with respect to the cosine distance, and then incorporated with the kernel-free nonlinear and nonparallel separating ideas to propose the FNQSSVM model by directly using two nonparallel quadratic surfaces for nonlinear classification. Computational results on three crawled real-life datasets in different domains show that the proposed FNQSSVM model outperforms the well-known and state-of-the-art classification methods in terms of classification accuracy for identifying helpful online reviews, within competitive computational time. The proposed method can be integrated into the decision support systems to assess the helpfulness of online reviews and facilitate the ranking of helpful reviews. Our findings can provide valuable managerial insights for online platforms, merchants and customers.