Detection And Recognition Of False Comments In Electronic Commerce

Comment data,as an important information data of e-commerce platform,plays an important role in business activities.However,the existence of many fake comments has brought erroneous orientation to consumers and business organizations,resulting in huge losses.Therefore,it is of great significance to detect and control them.There are a lot of comment data in e-commerce platform.Faced with so many comment data,the existing fake comment detection methods have some limitations.There are many kinds of commodities in e-commerce platform,and the types of comment data involved are complex.The classification features adopted by content-based methods are usually domain-dependent,and the classification performance depends on many correctly labeled comment data,so the generalization ability is poor.Although the method based on behavior analysis does not need to annotate the comment data,it relies on the specific user comment behavior,and the recognition rate is not high.In order to solve these problems,this paper proposes a systematic method to detect fake comments on e-commerce platform.The research content includes three aspects.One is to identify the target commodity containing fake comments;the other is to measure the similarity of the comment text;the third is to mine the features of fake comments recognition and build the detection model of fake comments.The main work of this paper is as follows:1)An algorithm for identifying the target product of fake reviews on e-commerce platform is proposed.E-commerce reviews cover a wide range of data,resulting in a decline in the accuracy of existing fake reviews detection methods.In order to obtain the sample data of fake reviews from large amount of e-commerce reviews and conduct targeted research,the identification of target commodities of fake reviews is studied first.It is found that the user rating behavior of commodities obeys the specific statistical law.When there is a certain amount of fake rating behavior,it will show a deviation from the normal rating behavior.By indexing the difference of seed delivery and using numerical indicators to identify the order of commodity lists,the higher the possibility of high-ranking commodities containing many fake comments.The experimental results show that the comments corresponding to TOP products ranked by this method do contain many fake comments,and this method can effectively identify the target goods containing fake comments by W.2)An algorithm for evaluating text similarity is proposed.In view of the low accuracy of traditional text similarity measurement methods,this paper constructs a tree structure of comment text by using the content organization characteristics of comment text,and decomposes the similarity measurement into similarity measurement between the corresponding tree layers,so that the objects of similarity measurement at each level are all words of the same type,and then calculates the similarity of each layer by using corresponding similarity measurement methods respectively.Finally,the similarity of each layer is fused according to the weight to get the overall similarity.The experimental results on real data sets show that the proposed method is more effective and accurate than other common measurement methods.3)A fake comment detection algorithm based on dynamic and static feature fusion is proposed.The existing fake comment detection methods do not make full use of the dynamic information contained in user's historical behavior.Firstly,the temporal analysis model is used to mine the dynamic features of user behavior from this dynamic information.Secondly,the dynamic features and static features at the user level are combined to find suspicious users,and the suspicious probability of users is propagated to the comments made by users to get the suspicious probability of comments.Finally,combining the suspicious probability of comments with static features at the comment level,learning strategies are used.A high-performance classifier is trained slightly to detect fake comments.Experiments on real data sets show that the performance of the proposed method is better than that of existing methods.