Research On Deceptive Reviews Detection Based On Multidimension Feature Weight

With the vigorous development of electronic commerce, online shopping has gradually become the main shopping behavior for people. Based on the characteristics of asymmetrical information generated between consumers and sellers during the online shopping process, the comments of online products are becoming very important basis for consumers when they are making decisions. Driven by the interests of the online sellers, however, more and more inveracious online comments break into the eyes of consumers, causing it becomes more and more difficult to obtain real network information for consumers. So, researches on identification of inveracious network comments are carried out. In recent years, with the continuous efforts of the vast number of researchers, different identification methods for inveracious network comments have been proposed. Though, these methods have their own advantages, the accuracy and efficiency of these methods can still be improved.Based on the previous researches, this thesis studies and proposes a new identification method for inveracious network comments which is based on the multi- feature weights of the comments. The proposed method uses the support vector machine(SVM) as the basic technique for text classification, takes the usefulness of network comments as a starting point, models the comments identification model by introducing the weights of the nine major features of the network comments, and finally identifies the inveracious network comments. Compared to the previous methods, the main advantage of our proposed method is that our method use a multi- feature weights of the comments for identification which can guarantee the accuracy and usefulness even though a single comment feature does not contribute obviously to the identification process. In this thesis, we carry out a series of experiments on the comments of cellphones collected from the Amazon website, and the experimental results show that the accuracy, recall and comprehensive classification rates of the network comments identification using SVM are higher when multi- feature weights of the comments are introduced. Then, the efficiency of our proposed method is verified.