Research Technology On Found Of Spammer Organizations Based On Multi-Feature Scale Space Model

The fake reviews on e-commerce sites may mislead consumers, affect the user’s purchasing decisions, but also may discredit the reputation of the business, seriously affect the normal business activities. Given the large number of fake reviews information often appears on the e-commerce sites, in order to protect the interests of consumers and businesses, we need to detect and treat the fake reviews information. The existing studies always identify the fake reviews according to the characteristics of the reviews’content and publishers’behaviors and the mostly training set are labeled. It’s a huge workload, and some samples are difficult to annotations by semantic analysis and simple behavioral characteristics analysis.To solve the above problem, we proposed the organization discovery technology based on a multi-feature scale space model. Taking into account the spammers are employed by the internet public relations firm, they earn profits through the publication of false information. We believe that the spammers for the same benefit, will be organized to co-publish fake reviews. To detect fake reviews, we first constructed the models of the Same Product Times, the Similarity of Rating, the Probability of Same Brand and processed the constructed reviews’user relationships network, gradually found the existed stable network organizations. Then we distinguished spammers by detecting the preferred key comments in the network organizations and then identify whether the organization’s network is a spammer organization. Eventually we speculated the posted fake reviews according to the spammers. Throughout the study, in order to find the spammer organizations, we analyzed the applicability of community discovery algorithm; in order to reduce the misjudgment that the single feature scale space model brings, we summarized the spammers’characteristics of behaviors, relationships and groups and built a multi-feature scale space model; in order to analyze the different characteristics of organizational changes in network scale space model structure process, we carried out the network structure deduction on multiple scales.Overall, the technology of finding spammer organizations based on the multi-feature scale space model proposed in this paper can greatly reduce the enormous labeled workload existed in the traditional fake reviews identification. According to different dimensions of feature selection based on comments can effectively identify the relationships between users, the multi-feature scale space can solve some difficulties that comments can not be labeled. Of course, some errors exist in the proposed method, and we particularly make some error analysis.