|
Category | Method | Data modeling | Candidate group discovery method | Spamming behavior evaluation | Detection method |
|
Group content and behavior analysis-based methods | Mukherjee et al. [3] | Construct behavioral and relation models | Frequent itemset mining | The average of spam indicators | Ranking |
Xu et al. [4] | Measure the pairwise similarity of groups | Frequent itemset mining | No | KNN-based method |
Xu and Zhang [7] | Use pairwise features to model the relations among colluders | No | The weighted sum of all the pairwise features | Ranking |
Xu and Zhang [8] | Use homogeneity-based behavior features to model the collusive review fraud | Frequent itemset mining | Similarity-based measure | Unified probabilistic model |
Zhang et al. [9] | Use a number of features to model the group spamming behavior | Frequent itemset mining | No | Semi-supervised classification |
|
Graph-based methods | Ye and Akoglu [10] | Model the dataset as a bipartite graph | No | Reviewers’ network footprint scores | Hierarchical clustering |
Choo et al. [11] | Model the dataset as a user relationship graph | No | No | Community-based method |
Wang et al. [12] | Model the dataset as a bipartite graph | Graph partitioning | The average of spam indicators | Ranking |
Do et al. [13] | Model the dataset as a review graph | No | The average of spam indicators | k-means clustering |
Do et al. [14] | Model the dataset as a review graph | No | The average of spam indicators | Fuzzy k-means clustering |
Han et al. [15] | Model the dataset as a user relationship graph | No | No | Graph spectrum analysis |
Wang et al. [16] | Model the dataset as a user relationship graph | Graph partitioning | The average of spam indicators | Ranking |
Cao et al. [21] | Model the dataset as a bipartite graph | No | The average of spam indicators | Hierarchical agglomerative clustering |
Zhang et al. [22] | Model the dataset as a user relationship graph | Label propagation | Linearly weighted sum of spam indicators | Ranking |
|