Research And Implementation Of Review Generation Algorithm Based On Deep Learning

Nowadays,more and more consumers buy the products or services on the consumer comment website or e-commerce platform.Such online platforms contain a large number of reviews written by consumers who purchase certain products or services,which can affect the decisions of consumers who are not familiar with the products or services.Driven by the importance of online reviews,malicious reviews emerge in an endless stream,resulting in more and more serious negative effects,which makes fake reviews become one of the important threats in the online evaluation system.With the development of text generation technology,review generation algorithm gradually replaces the expensive manual writing.Understanding the generation mechanism of fake reviews is helpful to design a more targeted detection algorithm.Not only that,in the fake review detection task,a large number of fake review dataset are often used as the basis of detection algorithm training.Therefore,how to generate confusing fake review is very important in the process of fake review detection.In this paper,we first design and implement a review generation algorithm based on deep learning.The algorithm uses both Transformer model and Gated Recurrent Unit(GRU)model,combined with their respective advantages and characteristics in processing text sequence,it can generate reviews with higher authenticity and novelty,thus providing better training data for fake review detection.Secondly,because of the shortcomings of the current automatic evaluation metrics in measuring novelty for generated text,a novel evaluation metrics DMet is proposed by us,which measures the diversity and novelty of the generated text by calculating word overlap,similarity penalty within the sequence and sentence length correction.Finally,this paper constructs an automatic generation and detection system of comment text.On the one hand,it can realize the end-to-end automatic generation of review text,on the other hand,it can be used as input to build an automatic evaluation and detection system of fake reviews.