基于深度学习的电子邮件分类系统的设计与实现

专业班级：       学生姓名：

指导教师：        职称：

摘要 本文设计并实现了一个基于深度学习的电子邮件分类系统，该系统采用卷积神经网络（CNN）和长短时记忆网络（LSTM）结合注意力机制（Attention）的算法，对电子邮件进行精准分类。系统开发工具采用PyCharm和Navicat，Web框架使用Django，前端框架则选用jQuery和layui。

系统首先构建了一个包含健康邮件和垃圾邮件的数据集，其中健康邮件存储于normal文件夹，数量约7000多份；垃圾邮件存储于spam文件夹，数量亦约7000多份。此外，系统还准备了一个包含约400份邮件的测试集，用于评估模型的性能。测试集中编号1-200的邮件为健康邮件，而编号7801-8000的邮件为垃圾邮件。

在模型训练阶段，本文先利用朴素贝叶斯算法编写了一个简单的垃圾邮件分类器，作为基准模型。随后，本文重点研究了基于深度学习的电子邮件分类方法，特别是CNN和LSTM-Attention的组合模型。该模型能够捕获邮件文本中的局部特征和时序依赖信息，并结合注意力机制对重要特征进行加权，从而提高分类的准确率。

实验结果表明，相较于朴素贝叶斯模型，基于深度学习的CNN和LSTM-Attention组合模型在电子邮件分类任务上表现出更优越的性能。本文的研究成果对于提升电子邮件分类系统的准确性和效率具有重要意义，有望在实际应用中发挥重要作用。

关键词：深度学习；电子邮件分类；卷积神经网络；长短时记忆网络；注意力机制

Design and implementation of an email classification system based on deep learning

Abstract This paper designs and implements a deep learning based email classification system. This system uses convolutional neural network (CNN) and long and short time memory network (LSTM) combined with attention mechanism (Attention) algorithm to accurately classify email. The system development tools use PyCharm and Navicat, the Web framework uses Django, and the front-end framework uses jQuery and layui.

The system first built a data set containing healthy messages and spam messages, with health messages stored in the normal folder with about 7000 copies and spam messages stored in the spam folder with about 7000 copies. In addition, the system has prepared a test set containing about 400 emails to evaluate the performance of the model. In the Test Set, numbers 1-200 were healthy, while numbers 7801-8000 were spam.

In the model training stage, this paper writes a simple spam classifier using naive Bayes algorithm as a benchmark model. Subsequently, this paper focuses on deep learning-based email classification methods, especially the combined model of CNN and LSTM-Attention. The model is able to capture local features and temporal dependence information in mail text, and weight important features with attention mechanism, thus improving the classification accuracy of classification.

Experimental results show that the deep learning-based combined CNN and LSTM-Attention model show superior performance on email classification task compared to the naive Bayes model. The research results of this paper are of great significance for improving the accuracy and efficiency of E-mail classification system, and are expected to play an important role in practical application.

Key words  deep learning; E-mail classification; convolutional neural network; long and short memory network; attention mechanism