Create Hugging Face compatible dataset format and update README

README.md

@@ -14,11 +14,12 @@ task_categories:
   - image-classification
 task_ids:
   - sentiment-classification
+  - multi-class-classification
 tags:
-  - covert advertisement
+  - advertisement
   - social-media
   - xiaohongshu
-  - RedNote
+  - redbook
 pretty_name: CHASM - Covert Advertisement on RedNote
 ---
 
@@ -34,45 +35,111 @@ A dataset containing posts from Xiaohongshu (RedNote) for text classification ta
 - **Language**: Chinese
 - **License**: MIT
 - **Dataset Size**: Contains two classes (label_0 and label_1)
+- **Splits**: Train (3089 samples), Validation (440 samples), Test (888 samples)
 
 ### Dataset Structure
 
+本数据集有两种格式：
+
+#### 1. 原始目录格式 (Original Directory Structure)
+
+数据按照以下目录结构组织：
+
+- `label_0/`: 非广告内容
+- `label_1/`: 广告内容
+
+每个样本有一个独立的文件夹（如 `train_889/`），包含：
+
+- `data.json`: 包含文本数据
+- 多个图像文件（JPG/WEBP 格式）
+
+#### 2. Hugging Face 格式 (Hugging Face Format)
+
+为了更好地与 Hugging Face 平台兼容，我们还提供了处理后的格式：
+
+- `hf_format/train.csv` 和 `hf_format/train.json`: 训练集
+- `hf_format/validation.csv` 和 `hf_format/validation.json`: 验证集
+- `hf_format/test.csv` 和 `hf_format/test.json`: 测试集
+- `hf_format/dataset_info.json`: 数据集元信息
+
+### 数据字段说明
+
 Each data sample contains the following fields:
 
-- `title`: Post title
-- `description`: Post description
-- `date`: Publication date and location
-- `comments`: List of comments
-- `image`: List of image filenames
-- `label`: Classification label (0 or 1)
+- `id`: 样本唯一标识符
+- `title`: 帖子标题
+- `description`: 帖子描述
+- `date`: 发布日期和位置
+- `comments`: 评论列表
+- `images`: 图像文件名列表
+- `label`: 分类标签 (0: 非广告, 1: 广告)
 
 ### Data Example
 
 ```json
 {
-    "title": "Sample Title",
-    "description": "Sample Description",
-    "date": "09-13 Location",
-    "comments": ["Comment 1", "Comment 2", ...],
-    "image": ["image1.webp", "image2.webp", ...],
+    "id": "train_889",
+    "title": "豆瓣9.2❗️几乎是全程震撼着读完的一本书❗️",
+    "description": "📚《人类新史》\n📝［美］大卫·格雷伯/［英］大卫·温格罗\n\n说实话我几乎是不读社科历史大部头的，一年能读上一本我都想夸自己厉害🤣所以看到这本书的厚度时心里真的发怵，担心自己读不下去，却没想到它出乎意料地好读，才读了百来页我就忍不住先来分享一波了，因为脑袋里信息量爆表了！！...",
+    "date": "编辑于 7 天前 湖南",
+    "comments": ["感谢推荐，已收藏", "你好厉害，能看英文版我看中文版才看了这么点就觉得很费劲", ...],
+    "images": ["1040g008317prti2fk45g48iu7h5portohqvdpdo!nd_dft_wlteh_webp_3_0.jpg", ...],
     "label": 0
 }
 ```
 
 ## Usage
 
+使用原始目录格式：
+
+```python
+import os
+import json
+from glob import glob
+
+# 遍历所有样本
+for label_dir in ['label_0', 'label_1']:
+    for sample_dir in glob(f"{label_dir}/*"):
+        if os.path.isdir(sample_dir):
+            # 读取数据文件
+            with open(os.path.join(sample_dir, 'data.json'), 'r', encoding='utf-8') as f:
+                data = json.load(f)
+
+            # 获取图像文件
+            images = glob(os.path.join(sample_dir, '*.jpg'))
+
+            # 处理数据...
+```
+
+使用 Hugging Face 格式：
+
+```python
+import pandas as pd
+
+# 加载数据
+train_df = pd.read_csv('hf_format/train.csv')
+val_df = pd.read_csv('hf_format/validation.csv')
+test_df = pd.read_csv('hf_format/test.csv')
+
+# 或者使用JSON格式
+import json
+with open('hf_format/train.json', 'r', encoding='utf-8') as f:
+    train_data = json.load(f)
+```
+
+使用 Hugging Face datasets 库：
+
 ```python
 from datasets import load_dataset
 
-# Load the dataset
+# 加载数据集
 dataset = load_dataset("Jingyi77/CHASM-Covert_Advertisement_on_RedNote")
 
-# View dataset information
+# 查看数据集信息
 print(dataset)
 
-# Access data
-print(dataset['label_0'][0])  # View first data point from label_0
-print(dataset['label_1'][0])  # View first data point from label_1
+# 访问数据
+print(dataset['train'][0])  # 查看训练集的第一个数据点
 ```
 
 ## Data Preprocessing

create_hf_dataset.py

@@ -0,0 +1,164 @@
+#!/usr/bin/env python3
+import os
+import json
+import pandas as pd
+from pathlib import Path
+from tqdm import tqdm
+
+def convert_to_hf_format():
+    """将现有目录结构转换为Hugging Face兼容的格式"""
+    print("开始转换数据集格式...")
+    
+    # 数据集路径
+    dataset_path = "/Users/zhengjingyi/Desktop/image_after"
+    
+    # 创建输出目录
+    output_dir = os.path.join(dataset_path, "hf_format")
+    os.makedirs(output_dir, exist_ok=True)
+    
+    # 为训练集、验证集和测试集创建空列表
+    train_data = []
+    val_data = []
+    test_data = []
+    
+    # 处理 label_0 目录
+    label_0_path = os.path.join(dataset_path, "label_0")
+    print("处理 label_0 目录...")
+    for folder in tqdm(os.listdir(label_0_path)):
+        folder_path = os.path.join(label_0_path, folder)
+        if not os.path.isdir(folder_path):
+            continue
+            
+        # 读取数据文件
+        data_file = os.path.join(folder_path, "data.json")
+        if not os.path.exists(data_file):
+            continue
+            
+        try:
+            with open(data_file, 'r', encoding='utf-8') as f:
+                data = json.load(f)
+                
+            # 获取图像文件列表
+            image_files = [f for f in os.listdir(folder_path) if f.endswith('.jpg') or f.endswith('.webp')]
+            
+            # 创建样本
+            sample = {
+                'id': folder,
+                'title': data.get('title', ''),
+                'description': data.get('description', ''),
+                'date': data.get('date', ''),
+                'comments': data.get('comments', []),
+                'images': image_files,
+                'label': 0
+            }
+            
+            # 根据目录名分配到相应的数据集
+            if folder.startswith('train'):
+                train_data.append(sample)
+            elif folder.startswith('val'):
+                val_data.append(sample)
+            elif folder.startswith('test'):
+                test_data.append(sample)
+        except Exception as e:
+            print(f"处理 {folder} 时出错: {str(e)}")
+    
+    # 处理 label_1 目录
+    label_1_path = os.path.join(dataset_path, "label_1")
+    print("处理 label_1 目录...")
+    for folder in tqdm(os.listdir(label_1_path)):
+        folder_path = os.path.join(label_1_path, folder)
+        if not os.path.isdir(folder_path):
+            continue
+            
+        # 读取数据文件
+        data_file = os.path.join(folder_path, "data.json")
+        if not os.path.exists(data_file):
+            continue
+            
+        try:
+            with open(data_file, 'r', encoding='utf-8') as f:
+                data = json.load(f)
+                
+            # 获取图像文件列表
+            image_files = [f for f in os.listdir(folder_path) if f.endswith('.jpg') or f.endswith('.webp')]
+            
+            # 创建样本
+            sample = {
+                'id': folder,
+                'title': data.get('title', ''),
+                'description': data.get('description', ''),
+                'date': data.get('date', ''),
+                'comments': data.get('comments', []),
+                'images': image_files,
+                'label': 1
+            }
+            
+            # 根据目录名分配到相应的数据集
+            if folder.startswith('train'):
+                train_data.append(sample)
+            elif folder.startswith('val'):
+                val_data.append(sample)
+            elif folder.startswith('test'):
+                test_data.append(sample)
+        except Exception as e:
+            print(f"处理 {folder} 时出错: {str(e)}")
+    
+    # 将数据保存为CSV和JSON格式
+    print("保存数据...")
+    
+    # 创建DataFrame
+    train_df = pd.DataFrame(train_data)
+    val_df = pd.DataFrame(val_data)
+    test_df = pd.DataFrame(test_data)
+    
+    # 保存为CSV
+    train_df.to_csv(os.path.join(output_dir, "train.csv"), index=False)
+    val_df.to_csv(os.path.join(output_dir, "validation.csv"), index=False)
+    test_df.to_csv(os.path.join(output_dir, "test.csv"), index=False)
+    
+    # 保存为JSON
+    with open(os.path.join(output_dir, "train.json"), 'w', encoding='utf-8') as f:
+        json.dump(train_data, f, ensure_ascii=False, indent=2)
+    
+    with open(os.path.join(output_dir, "validation.json"), 'w', encoding='utf-8') as f:
+        json.dump(val_data, f, ensure_ascii=False, indent=2)
+    
+    with open(os.path.join(output_dir, "test.json"), 'w', encoding='utf-8') as f:
+        json.dump(test_data, f, ensure_ascii=False, indent=2)
+    
+    # 创建索引文件
+    with open(os.path.join(output_dir, "dataset_info.json"), 'w', encoding='utf-8') as f:
+        info = {
+            "name": "CHASM-Covert_Advertisement_on_RedNote",
+            "version": "1.0.0",
+            "description": "A dataset containing posts from Xiaohongshu (RedNote) for text classification tasks, specifically focused on identifying covert advertisements",
+            "splits": {
+                "train": {
+                    "num_examples": len(train_data)
+                },
+                "validation": {
+                    "num_examples": len(val_data)
+                },
+                "test": {
+                    "num_examples": len(test_data)
+                }
+            },
+            "features": {
+                "id": {"dtype": "string", "description": "Unique ID for the sample"},
+                "title": {"dtype": "string", "description": "Post title"},
+                "description": {"dtype": "string", "description": "Post description"},
+                "date": {"dtype": "string", "description": "Publication date and location"},
+                "comments": {"sequence": {"dtype": "string"}, "description": "List of comments"},
+                "images": {"sequence": {"dtype": "string"}, "description": "List of image filenames"},
+                "label": {"dtype": "int32", "description": "Classification label (0 or 1)"}
+            }
+        }
+        json.dump(info, f, ensure_ascii=False, indent=2)
+    
+    print(f"转换完成！数据已保存到 {output_dir}")
+    print(f"训练集: {len(train_data)} 样本")
+    print(f"验证集: {len(val_data)} 样本")
+    print(f"测试集: {len(test_data)} 样本")
+
+if __name__ == "__main__":
+    convert_to_hf_format()