Recognizing Deceptive Reviews Based on Weighted Multi-Instance Unbalanced Support Vector Machine

Recognition performance of detecting deceptive reviews model depends heavily on quality of manual annotated training data. However, manual label of reviews often has misclassification. Considering characteristics of training data that deceptive reviews are less than true ones and misclassified reviews in training data, a model of recognizing deceptive reviews based on weighted multi-instance unbalanced support vector machine is proposed. Theoretical analysis and experimental results showed detection model effectively reduce impact of misclassified data in training data and improve recognition performance of deceptive reviews in unbalanced reviews dataset.