Navyasri17/phishing_emails-data

# 🛡️ Phishing Email Classification Dataset This dataset is curated for fine-tuning LLMs on the task of phishing email detection. It originates from [this Kaggle dataset](https://www.kaggle.com/datasets/subhajournal/phishingemails) and has been transformed to better suit LLM-based classification tasks. ## 📦 Dataset Features - Each row is a labeled email, with either: - `safe email` (label = 0) - `phishing email` (label = 1) - The dataset includes metadata (sender, receiver, date, subject) and cleaned email body. - Two main columns: - `Email Text`: Complete formatted text including metadata and message content. - `label`: Binary label indicating if the email is phishing. ## 🧠 LLM Fine-Tuning Ready Processed using a `phishing_items.py` parser: - Truncates or filters emails based on token limits for LLM input (between 30 and 250 tokens). - Builds classification prompts in the format: ``` Is the following email safe or phishing?? [email content] Email type is: [safe email/phishing email] ``` - Optimized for models such as `meta-llama/Meta-Llama-3.1-8B`. ## 🧼 Preprocessing Highlights - Removes non-informative characters (e.g., `=`, `>`, `\`) and extra whitespace. - Tokenized with Hugging Face's `AutoTokenizer`. - Discards overly short emails (under 120 characters or under 30 tokens). ## 🗂️ Example Usage ```python from phishing_items import Item item = Item(data_row) if item.include: print(item.prompt) ``` ## 📚 Source - Original dataset: [Kaggle - Phishing Emails](https://www.kaggle.com/datasets/subhajournal/phishingemails) - Transformed by: [your GitHub or Hugging Face handle]# 🛡️ 钓鱼邮件分类数据集（Phishing Email Classification Dataset） 本数据集专为大语言模型（Large Language Model, LLM）的钓鱼邮件检测微调任务打造，其原始数据源自[该Kaggle数据集](https://www.kaggle.com/datasets/subhajournal/phishingemails)，并经过适配基于大语言模型的分类任务的优化处理。 ## 📦 数据集特征 - 每一行均为一条带标签的邮件，标签分为两类： - 正常邮件（`safe email`，标签值为0） - 钓鱼邮件（`phishing email`，标签值为1） - 数据集包含元数据（发件人、收件人、发送日期、主题）与清洗后的邮件正文。 - 核心字段共两项： - `Email Text`：包含元数据与邮件正文的完整格式化文本。 - `label`：二元标签，用于标识该邮件是否为钓鱼邮件。 ## 🧠 适配大语言模型微调 本数据集通过`phishing_items.py`解析脚本完成预处理： - 根据大语言模型输入的Token限制对邮件进行截断或过滤（Token数介于30至250之间）。 - 采用如下格式构建分类提示： 以下邮件属于正常邮件还是钓鱼邮件？ [邮件内容] 该邮件类型为：[正常邮件/钓鱼邮件] - 针对`meta-llama/Meta-Llama-3.1-8B`等模型进行了优化。 ## 🧼 预处理要点 - 移除无意义字符（如`=`、`>`、``）与多余空格。 - 使用Hugging Face的`AutoTokenizer`进行Token化处理。 - 过滤过短的邮件（字符数少于120或Token数少于30）。 ## 🗂️ 示例用法 python from phishing_items import Item item = Item(data_row) if item.include: print(item.prompt) ## 📚 来源 - 原始数据集：[Kaggle - 钓鱼邮件数据集](https://www.kaggle.com/datasets/subhajournal/phishingemails) - 转换处理方：[你的GitHub或Hugging Face账号标识]