基于对抗迁移学习模型的环境类虚假投诉检测

Environmental False Complaint Detection Based on Adversarial Transfer Learning Model

为实现环境类虚假投诉举报检测,提出一种基于对抗迁移学习方法的虚假投诉举报检测模型。首先,以长短期记忆(long-short term memory,LSTM)网络为特征抽取器抽取微博谣言(源域)和投诉举报文本(目标域)的共享特征;然后,使用对抗学习方法进行领域适配,将源域特征和目标域特征进行特征对齐;最后,由分类器输出分类结果,并由分类损失和领域适配损失共同更新网络参数。通过模型对比实验和消融实验可知,模型的F_(1)达到了79.61%。结果表明,对抗迁移学习模型具有较好的性能,适合应用在环境类虚假投诉举报检测任务中。

To achieve environmental false complaint report detection,a false complaint and report detection model was proposed based on adversarial transfer learning method.First,a long-short term memory(LSTM) network was used as a feature extractor to extract the shared features of Weibo rumors(source domain) and complaint report text(target domain).Second,the domain adaptation was performed by using the adversarial learning method to align the source domain features with the target domain features.Finally,the classification results were output by the classifier,and the network parameters were updated by the classification loss and the domain adaptation loss.Model comparison experiments and ablation experiments were designed,and the F_(1) value of the model reached 79.61%,indicating that the adversarial transfer learning model has good performance and is suitable for application in the the task of detecting environmental false complaints and reports.