Fake Review Detection Based On Deep Learning

With the rapid development of the Internet,online shopping has replaced the store shopping with its convenience and quick advantage,and it has become the most popular consumption mode.For the purpose of profit,there are more and more fake reviews in the e-commerce platform.This misleads the consumer's buying trend to a certain extent,causing the consumer to buy goods that are not consistent with the description,and gradually lose trust in the e-commerce platform.In order to purify the e-commerce environment,researchers use traditional classifier to detect fake reviews based on the various features of commentators and commentators.Although these methods have achieved certain results,the process of extracting features depends on the knowledge of experts,and does not consider the features of the related product.These methods cannot be widely used.Generally speaking,fake reviews use more sentimental words to describe the features of the target product than actual reviews.Spammers use different words to describe different products.In order to make the detection method applicable to different fields,two kinds of fake review detection methods based on deep learning are proposed by using product related features.The first is using product related features to detect fake review.This method proposes a novel convolutional neural network model to integrate the product related features through a product word composition model.To reduce overfitting,a bagging model is introduced to bag the neural network model with two efficient classifiers.The second is fake review detection based on FastText.In this method,the Word2Vec model is trained.Based on this model,the product feature glossary is extended,and the text vector is established.After extracting features using coiling layer and pooling layer,FastText is applied to classify.Using this method,imbalanced class problem is solved and the training speed of the model is improved.In order to verify the effectiveness of algorithms proposed in this paper,we use Python language to crawl the product reviews data set of Amazon website,and design several group comparison experiments for the above two algorithms based on this dataset.The result show that the first algorithm use product related features can improve the accuracy of fake review detection,and using bagging model can reduce overfitting.The second algorithm show that Word2Vec can better express the deep semantics of text,and using FastText can solve imbalanced class problem and shorten the training time of the model in the condition of large data.