Research On E-commerce Fake Review Identification Method Based On Deep Learning

With the further popularization of mobile internet and intelligent devices,ecommerce has rapidly developed and penetrated into people’s daily lives.Consumers increasingly rely on online reviews in their consumption decisions.False reviews have become an important tool for illegal businesses to seek benefits through information asymmetry.The proliferation of false comments not only destroys the fair competition order in the e-commerce market,but also has a negative impact on consumers’ shopping experience and trust in e-commerce platforms.Therefore,false reviews have become a hot topic in e-commerce research.However,existing false comment recognition models cannot extract comment text features well,nor fully consider the correlation between comment text features and behavioral features.The recognition effect of false comments needs to be improved.In this regard,this article combs and analyzes the literature through the knowledge map analysis software Citespace,and proposes a false review model based on the triple convolution twin network and CatBoost,CNN-Tri BERT-CatBoost.Firstly,external information is introduced through the BERT pre training model to generate a comment vector that reflects context information.Then,based on the twin network,we propose a triplet convolutional twin network to enhance the semantic expression ability of word vectors.By improving the similarity between word vectors with the same label,and reducing the similarity between word vectors with different labels,we can enhance the recognition performance of subsequent models from the perspective of semantic similarity.Finally,the CatBoost model is used to combine text features and behavioral features and classify them to identify false comments.The results of comparative experiments on the Volkswagen Review dataset show that the F1,AUC,accuracy,accuracy,and recall rates of the model in this paper reach0.7961,0.9273,0.8620,0.8528,and 0.7469,respectively.The best of the four evaluation indicators is obtained in the comparative model.Compared to existing deep learning models such as Text CNN,Text RNN,attention Text RNN,Text RCNN,Fast Text,DPCNN,and Transform,the F1 value has increased by 7.6,16.3,5.2,3.2,7.9,5.0,and 6.8percentage points,respectively.In addition,this article also proves the effectiveness of the improved model by comparing models with different structures.Figure [32] Table [12] Reference [91]...