Fake Review Detection Based On Sentiment Analysis

With the development of the Internet industry,a large number of users buy products and services online.Users can write their feelings in the comments to express the value and quality of the goods and services purchased.The credibility of online comments greatly affects the reputation and economic benefits of enterprises.So online sellers hire people to write fake reviews to recommend their own products or denigrate competitors’ products.But with new types of online scams like part-time brushing popping up,fake review detection has become a challenging task.Among the existing methods for detecting fake comments,many researchers have introduced text emotion into fake comments,but they only take text emotion as one of the text features,or identify fake comments by the consistency of text emotion and user rating,without in-depth study on the relationship between text emotion and fake comments.On this basis,this paper proposes a fake comment detection method based on sentiment analysis,which has three steps in total.The first step is to use the Text Blob method to conduct sentiment analysis on the comment text,dig the expression of the word information in the sentence,judge the emotional tendency,calculate the precise emotion score,observe the relationship between emotion score and user rating,which can be divided into four different combinations(high user rating and high emotion score combination,high user rating and low emotion score combination,high user rating and low emotion score combination,high user rating and high emotion score combination,high user rating and low emotion score combination,high user rating and high emotion score combination,high user rating and low emotion score).Low user rating combined with high emotion score,low user rating combined with low emotion score);In the second step,text features and behavioral features of fake reviews are selected to integrate the features of fake reviews.In view of the nature of product review data,the number of fake reviews is far less than that of real reviews.In order to balance the data set,cross-check method and downsampling method are used to construct the training set and test set respectively for the four combinations.In the third step,a supervised machine learning algorithm is used to obtain the optimal index of each combination by adjusting the algorithm parameters,and then the model is compared with the original data set to prove the effectiveness of the model.At the same time,the importance of the emotion of the comment text to the detection of false comments is judged.This paper innovatively divides emotion scores and user ratings into four different combinations to detect fake reviews.The experimental results show that the Text Blob method is more objective than the Vader Sentiment method,has better effect of sentiment analysis,and is better applied to the judgment and calculation of the sentiment inclination of comments.There are some differences between the fake comments and the real comments in the review text.The fake comments contain more emotional words,and the tone is strong and the wording is too comprehensive.Emotion-related language feature extraction is more effective for the detection of false comments,and text comments with too extreme expression are more likely to detect false comments.