| In recent years,with the rapid development of e-commerce websites and social networks,comment information has become an important factor in determining whether people consume.However,the existence of review spammer groups affects the authenticity and fairness of the e-commerce platform.In order to detect review spammer groups,various studies have been conducted.The common problem with these methods is that review spammer groups cannot be well identified,resulting in low detection accuracy.This topic focuses on the limitations of the current research and has conducted research from the following two aspects.The specific contents are as follows.First of all,to solve the problem that the existing algorithms can not fully tap the potential relationship between users and do not divide the detection indicators of importance,we propose a review spammer groups detection algorithm based on metagraphs.This algorithm by extracting information such as the review times and ratings of users of products,a meta-graph-based heterogeneous network is constructed to mine potential relationships between users.Then,the improved Deepwalk algorithm is used to walk to obtain the low-dimensional vector representation of the user node,with the candidate group is obtained by combining the Canopy algorithm and the K-Means algorithm for clustering.Finally,the dynamic weight of each detection index is calculated according to the entropy value method,via the detection indexes after the weighting are fused to obtain group suspicion and sort it,thereby identifying the review spammer groups.Secondly,to solve the problem that the existing algorithms do not mine relevant attributes in product review information sufficiently,a review spammer groups detection algorithm based on feature fusion is proposed.The algorithm fusion of product with review time,product with rating produce two new attributes,construct a heterogeneous network based on feature fusion after the user and the new attribute are merged,by use the idea of node embedding to get the user’s vector representation.After that,the user vector is input into the convolutional neural network for training,so the clustering is performed in stages according to the user prediction value,and the user tightness in the candidate group is calculated to divide the review spammer groups to obtain the final detection result.Finally,for the algorithm proposed in this paper,experiments are performed on the Amazon and Miami data sets,the experiments are compared with existing algorithms also.The experimental results show that the accuracy rate and recall rate of the proposed algorithm on a given data set has certain advantages. |
