A Cross-modal Detection Approach On Hidden Astroturfing

The purpose of this paper is to do the detection of implicit astroturfing based on cross-modal.Discovering new clues for the identification of implicit astroturfing.We proposed two methods for the detection of implicit astroturfing,one is a method called CDM which can do detection of implicit astroturfing based on cross-modal,and the other one is a method called FDM which can apply fast detection of implicit astroturfing.CDM found and researched two new characteristics of implicit astroturfing:1)the text comments and picture descriptions given by users are very similar to each other;2)the picture comments uploaded have nothing to do with the commodities,so we realize the cross-modal detection of the implicit astroturfing based on these characteristics.And the experiment shows that CDM has a high accuracy in the test data set.FDM discovered new clues of implicit astroturfing-user comments emotion is inconsistent with the rating,and we realized the rapid detection of implicit astroturfing based on the sentiment analysis of user comments.And the experiment shows that FDM can improve the processing efficiency of the algorithm under the premise of ensuring the accuracy of emotion classification.Based on the two methods proposed above,an implicit astroturfing detection prototype system based on emotion analysis has been designed and implemented,and software copyright has been obtained.The main contributions of this work are:1)discovering new clues of implicit astroturfing,and implicit astroturfing detection have been implemented for texts which inconsistent images and emotion;2)marking the sample of implicit astroturfing,which realized the construction of about 10,000 sample bases and make a well preparation for the subsequent work based on discriminant learning;3)designing an implicit astroturfing detection system based on emotion analysis,we obtain the soft copyright,and cooperate with the National Science Foundation of China.