Fake Reviews Recognition Based On Multi-model Feature Extraction And Fusion

With the rapid development of Internet technology,more and more users are actively participating in the evaluation of various virtual service platforms and APPs,such as comments or scoring.These kinds of information play a vital role in the decision-making of new users,which in turn will directly affect the economic benefits of virtual service platforms and products.Massive text information is of great value to various APPs service providers and their users,but some businesses or users inevitably provide fake consumer experience in the reviews or comments for their benefits,and even imitate the language style of real reviewers,forming fake reviews and misleading users to judge the authenticity of the product.Therefore,fake reviews recognition are very important for extracting their true value.Accordingly,the paper studies fake reviews recognition based on multi-model feature extraction and fusion to possibly improve the accuracy.Based on the review of the current research work on fake reviews recognition,the paper proposes the following research contents:(1)Fake reviews recognition strategy based on traditional multi-model feature extraction and fusion: Aiming at the problems of incomplete representation of comment texts features and incomplete fused features,using commonly used traditional text feature extraction models to first extract multiple features as the evaluation text contents,word frequency,word vector,document and deep semantic.The emotional tendency of fake review is further analyzed to represent the language features.Based on semantic features and language features,three fusion strategies are then proposed for realizing the multi-level integration of features extracted with the multi-model.The fused feature is adopted to train random forest and Light GBM as the recognition model.The proposed algorithm is applied to the hotel comments set,and the results empirically demonstrate its effectiveness on accurately find fake reviews.(2)Fake reviews recognition based on feature fusion with multi-layer attention mechanism: Content(1)cannot sufficiently extract the semantic information by simply using traditional models,and therefore,a bidirectional LSTM neural network and attention mechanism are further studied to powerfully obtain more informative features.The bidirectional LSTM is constructed to get the untrustworthy key text of the evaluation,and the attention mechanism is introduced to obtain important semantic information related to the fake reviews.In the feature fusion,the self-attention mechanism is used to further enhance the semantic information of the original comment texts to improve the semantic representation ability.The performance of the proposed algorithm on more accurately identifying the fake reviews is experimentally illustrated by applying it to the golden comments and hotel evaluation one.This thesis contains 21 figures,10 tables and 95 references.