Research On Recognition Of Fake Transaction Behavior In Online-shopping From The Perspective Of Regulators

China's online retail salesincreased to 3 trillion and 880 billion yuan in 2015(a 39.07%rise),which was equivalent to 12.89%of the total retail sales of China during the same period.This channel is an important supplement to the traditional channel,improving transaction efficiency andallocation of social resources.At the same time,the information asymmetry,which also exists in the traditional market,is moreserious in theonline market for its geographical isolation.Online shopping platforms developed credite valuation system,customer review system and so on to protect buyers' equities.But some sellers found they couldincrease their rankings,traffic,conversion and packingpoor quality products into quality products for sale with fake transactions in these kinds of systems.It is thephenomenon that this paper concentrates on.The damage caused by this phenomenon could be summarized asbuyers's damage,sellers'loss,resource ofonline shopping platform flowing out,negative impacts on the transaction efficiency and the allocation of social resources.This research designs a recognition system for fake transactions with the viewpoint of regulators,which is proper for this phenomenon.Analyzingthe major members of this market behavior(buyers,sellers,platforms and regulators)by quality control system and cause-effect diagram,this research concentrates on responsible govenors to cope with this phenomenon.This research makes some efforts to recognize fake transactions.For instance,this paper designs a comprehensivesystem that combined by information acquisition,text analysis,and the identification of Outliers and Abnormal Trends.In terms of information acquisition,this system takes some information into account likethe fram of given platform,detail information requirement for further research(like timeline,user name,sales,evaluation and evaluation content).After this part of consideration,this research designslogic architecture withinthe Octopus Spiders to get required informations.After the information acquisition,this research starts to carry out the subsequent identification measures design.Text analysis includes two processes:text mining and descriptive statistics.The process of text mining conducts segmentation,clustering,word frequency statistic and extracting emotional tendency from reviews,calculating reviews,emotional scores,which makes the horizontal comparison possible.Descriptive statistics carries out words and sentences analysisto determine suspicious commodities for the final stage of analysis.In particular,words per review and the rate of positive and negative words that make up the mainly indicators for the words analysis.At the same time,horizontal comparisons are made according to emotional score of the review.The suspicious communities are analysed by statistics methods of anomaly detection.There are distinct differences between general sales and fake transactions,so the two processes are defined as the normal state and the abnormal state.Sellers carry out fake transactionsto increase sales,evolutions and customers' reviews,hence the previous phenomenon could be identified by corresponding abnormal conditions.The following measures can be divided into two phases:text analysis and outlier identification.suspicious commodities are analyzedas follows:identifying outliers during sales by standard deviation and boxplot technology,then verifying the outliers with ARIMA model;finding anomaly trends of reviews with trend simulation,them verifying the result by outliers' identification;discovering abnormal phenomenon of reviews,then verify abnormal trendswithholt trend exponential smoothing.After identifying principles and designing measure,this papertrials in M shopping platform with its public information.The results prove that the effectiveness of measures is apparent.The test identifies two suspicious commodities that should be further supervised by governors.This research partly makes up for gaps of fake transaction in online shopping recognition field and makes some corresponding contribution to improve the quality control system.In the 5th chapter,the summary and adviceis made and two future directions,which is composed of machine learning and text mining,are presentedfor these kinds of research.