Research On The Method Of Identifying Unrealistic Commentators Based On Shopping Platform Comments

Major e-commerce platforms are rapidly rising,and online shopping has gradually become a mainstream shopping method.Users are easily affected by online commodity reviews when shopping online.Driven by huge sales profits,many unrealistic commentators promote their products by publishing false comments or publishing biased comments to harm the interests of competitors and at the same time cause losses to buyers.In order to achieve a relatively fair trading environment,identifying unrealistic commentators has become a research hotspot in the field.The unrealistic commentators identified in this paper refers to the reviewer who is employed by the merchant or platform to publish comments that do not conform to the real attribute of the commodity.Previous research methods mainly used a single method to identify it,and the accuracy of identification is usually not high,which is difficult to apply to daily practice.In order to identify the sender of unrealistic commentators,the main work of this paper is to propose a hybrid recognition method(HY_RE)that combines text similarity,text semantics and user behavior characteristics.Firstly,this paper calculates the similarity of comment texts on the basis of the existing improved vector space model(VSM_C).Secondly,text semantics is analyzed through text-topic model(LDA),and the deviation degree of comment text-topic is calculated.Then define and calculate the relevant user behavior characteristics(USER_C)such as reviewer account level,comment publishing time,comment times,etc.Finally,the regression equation is used to integrate the calculation results of the previous steps,and a better identification result is obtained.In order to verify the effectiveness of the recognition method,this paper defines three evaluation indexes of accuracy,recall rate and f-measure,and compares the proposed method with the single recognition method,which verifies that the proposed method is obviously better than the single recognition method in the comprehensive evaluation index f-measure value.