基于对比学习的零样本对象谣言检测
|
陈珂,张文浩
|
Zero-shot object rumor detection based on contrastive learning
|
Ke CHEN,Wenhao ZHANG
|
|
表 4 不同方法在Zeo-Weibo数据集上的实验结果 |
Tab.4 Results of different methods on Zeo-Weibo object rumor detection dataset % |
|
方法 | Avg-F1 | Avg-R | Avg-P | Avg-A | SCO | CNN[33] | 68.82 | 76.43 | 71.71 | 82.97 | 74.98 | BiLSTM[34] | 70.68 | 72.02 | 70.66 | 76.59 | 72.48 | CNN-BiLSTM[35] | 64.01 | 66.58 | 62.84 | 75.88 | 67.32 | Arc1[22] | 79.64 | 80.26 | 79.73 | 81.58 | 80.30 | Arc2[22] | 76.29 | 77.99 | 76.97 | 82.39 | 78.41 | Arc3[22] | 74.77 | 77.33 | 76.40 | 82.02 | 77.63 | BERT[29] | 85.24 | 87.09 | 83.90 | 85.39 | 85.40 | PT-HCL[12] | 85.35 | 85.09 | 86.39 | 84.88 | 85.43 | ZPTHCL (本研究) | 87.58 | 87.25 | 88.24 | 87.13 | 87.55 |
|
|
|