False Complaint Report Detection Based on Contrastive Learning Model

To achieve fake complaint reporting detection, we propose a false complaint detection model based on contrastive learning. Firstly, the pre-trained model is used as a feature extractor to extract the complaint reporting sample features; secondly, the sample features of different categories are compared and learned to reconstruct the feature distribution in the feature space; finally, the classifier outputs the classification result and the network parameters are updated by the contrastive loss and balance loss together. In this paper, we designed comparison experiments and ablation experiments, and the F1 and AUC values of the model reached 81.40% and 81.46%, indicating that the contrastive learning model has better performance and is suitable to be applied in the fake complaint detection task.