Update README.md
Browse files
README.md
CHANGED
|
@@ -8,4 +8,68 @@ tags:
|
|
| 8 |
size_categories:
|
| 9 |
- n<1K
|
| 10 |
---
|
| 11 |
-
基于CCF23-EVAL任务6的电信网络诈骗案件数据集合成了风险短信与对话数据集，并基于多样性、真实性和是否满足人类偏好进行筛选，可用于风险细粒度分类任务和风险摘要生成任务测评。
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 8 |
size_categories:
|
| 9 |
- n<1K
|
| 10 |
---
|
| 11 |
+
# 基于CCF23-EVAL任务6的电信网络诈骗案件数据集合成了风险短信与对话数据集，并基于多样性、真实性和是否满足人类偏好进行筛选，可用于风险细粒度分类任务和风险摘要生成任务测评。
|
| 12 |
+
|
| 13 |
+
### 短信生成数据集筛选前后的评价结果比较
|
| 14 |
+
|
| 15 |
+
| 数据集 | 数据评估指标 | 案例生成方式 | 属性提示生成方式 | 案例生成方式 | 属性提示生成方式 |
|
| 16 |
+
|------------|----------------------|--------------|------------------|--------------|------------------|
|
| 17 |
+
| | 余弦相似度↓ | | | | |
|
| 18 |
+
| 所有类别 | | 0.7149 | **0.6943** | 0.7041 | **0.6704** |
|
| 19 |
+
| 冒充电商物流客服类 | | 0.7542 | **0.6981** | 0.7606 | **0.7331** |
|
| 20 |
+
| 虚假网络投资理财类 | | 0.7967 | **0.7120** | 0.8029 | **0.7108** |
|
| 21 |
+
| 虚假信用服务类 | | 0.7840 | **0.7050** | 0.7738 | **0.7010** |
|
| 22 |
+
| 虚假购物、服务类 | | 0.7088 | **0.6931** | 0.6879 | **0.6672** |
|
| 23 |
+
| 冒充公检法及政府机关类 | | 0.7979 | **0.7088** | 0.7835 | **0.6935** |
|
| 24 |
+
| 冒充领导、熟人类 | | 0.7765 | **0.7063** | 0.7540 | **0.7132** |
|
| 25 |
+
| 网络婚恋、交友类 | | 0.7469 | **0.6972** | 0.7904 | **0.7383** |
|
| 26 |
+
| 冒充军警购物类诈骗 | | 0.8091 | **0.6913** | **0.7877** | 0.7942 |
|
| 27 |
+
| 网黑案件 | | 0.8019 | **0.6989** | 0.8847 | **0.7993** |
|
| 28 |
+
| 无风险 | | 0.7231 | **0.6965** | 0.7029 | **0.6758** |
|
| 29 |
+
| | 欧氏距离↑ | | | | |
|
| 30 |
+
| 所有类别 | | 0.6968 | **0.7184** | 0.7063 | **0.7413** |
|
| 31 |
+
| 冒充电商物流客服类 | | 0.6544 | **0.7148** | 0.7251 | **0.7445** |
|
| 32 |
+
| 虚假网络投资理财类 | | 0.5992 | **0.7004** | 0.6191 | **0.7195** |
|
| 33 |
+
| 虚假信用服务类 | | 0.6186 | **0.7077** | 0.6549 | **0.6986** |
|
| 34 |
+
| 虚假购物、服务类 | | 0.7037 | **0.7196** | 0.6050 | **0.6736** |
|
| 35 |
+
| 冒充公检法及政府机关类 | | 0.5961 | **0.7034** | **0.6136** | 0.6041 |
|
| 36 |
+
| 冒充领导、熟人类 | | 0.6251 | **0.7061** | 0.4655 | **0.6005** |
|
| 37 |
+
| 网络婚恋、交友类 | | 0.6617 | **0.7157** | 0.7099 | **0.7365** |
|
| 38 |
+
| 冒充军警购物类诈骗 | | 0.5812 | **0.7214** | 0.6439 | **0.6773** |
|
| 39 |
+
| 网黑案件 | | 0.5912 | **0.7139** | 0.5921 | **0.7023** |
|
| 40 |
+
| 无风险 | | 0.6875 | **0.7161** | 0.6317 | **0.7112** |
|
| 41 |
+
| | 任务相关性↑ | **0.5665** | 0.5159 | **0.7775** | 0.7245 |
|
| 42 |
+
| | 人类偏好评分↑ | **1.5534** | 0.8722 | **1.9457** | 1.5478 |
|
| 43 |
+
|
| 44 |
+
---
|
| 45 |
+
|
| 46 |
+
### 对话生成数据集筛选前后的评价结果比较
|
| 47 |
+
|
| 48 |
+
| 数据集 | 数据评估指标 | 案例生成方式 | 属性提示生成方式 | 案例生成方式 | 属性提示生成方式 |
|
| 49 |
+
|------------|----------------------|--------------|------------------|--------------|------------------|
|
| 50 |
+
| | 余弦相似度↓ | | | | |
|
| 51 |
+
| 所有类别 | | 0.7910 | **0.7592** | 0.7456 | **0.7049** |
|
| 52 |
+
| 冒充电商物流客服类 | | 0.8172 | **0.7854** | 0.7966 | **0.7583** |
|
| 53 |
+
| 虚假网络投资理财类 | | 0.8452 | **0.8034** | 0.8134 | **0.7487** |
|
| 54 |
+
| 虚假信用服务类 | | 0.8452 | **0.8074** | 0.8099 | **0.7799** |
|
| 55 |
+
| 虚假购物、服务类 | | 0.7893 | **0.7618** | 0.7564 | **0.7238** |
|
| 56 |
+
| 冒充公检法及政府机关类 | | 0.8678 | **0.7966** | 0.8261 | **0.7515** |
|
| 57 |
+
| 冒充领导、熟人类 | | 0.8562 | **0.7744** | 0.7645 | **0.7525** |
|
| 58 |
+
| 网络婚恋、交友类 | | 0.8225 | **0.8260** | 0.7660 | **0.7912** |
|
| 59 |
+
| 冒充军警购物类诈骗 | | 0.8815 | **0.8594** | 0.8560 | **0.7843** |
|
| 60 |
+
| 网黑案件 | | 0.8745 | **0.8047** | 0.8224 | **0.7756** |
|
| 61 |
+
| 无风险 | | 0.7887 | **0.7672** | 0.7382 | **0.7115** |
|
| 62 |
+
| | 欧氏距离↑ | | | | |
|
| 63 |
+
| 所有类别 | | 0.6093 | **0.6486** | 0.6651 | **0.7085** |
|
| 64 |
+
| 冒充电商物流客服类 | | 0.5741 | **0.6161** | 0.6018 | **0.6507** |
|
| 65 |
+
| 虚假网络投资理财类 | | 0.5306 | **0.5913** | 0.5799 | **0.6622** |
|
| 66 |
+
| 虚假信用服务类 | | 0.5316 | **0.5849** | 0.5851 | **0.6241** |
|
| 67 |
+
| 虚假购物、服务类 | | 0.6119 | **0.6455** | 0.6532 | **0.6886** |
|
| 68 |
+
| 冒充公检法及政府机关类 | | 0.4893 | **0.6015** | 0.5618 | **0.6587** |
|
| 69 |
+
| 冒充领导、熟人类 | | 0.5098 | **0.6298** | 0.6416 | **0.6577** |
|
| 70 |
+
| 网络婚恋、交友类 | | **0.5652** | 0.5598 | **0.6416** | 0.6099 |
|
| 71 |
+
| 冒充军警购物类诈骗 | | 0.4664 | **0.5061** | 0.5140 | **0.6183** |
|
| 72 |
+
| 网黑案件 | | 0.4776 | **0.5910** | 0.5653 | **0.5938** |
|
| 73 |
+
| 无风险 | | 0.6122 | **0.6387** | 0.6737 | **0.7021** |
|
| 74 |
+
| | 任务相关性↑ | **0.5860** | 0.5577 | **0.7704** | 0.7498 |
|
| 75 |
+
| | 人类偏好评分↑ | 1.1921 | **1.5272** | 1.7836 | **1.9586** |
|