基于Fuzzy Bayesian的学习和推理方法在邮件系统中的应用

发布时间：2018-07-04 21:16

本文选题：垃圾邮件过滤 + 朴素贝叶斯　；参考：《中国地质大学(北京)》2017年硕士论文

【摘要】：近年来,随着网络技术的迅速发展,越来越多的人选择通过电子邮件(E-mail)这种简易而又快捷的通信方式来与他人联络。但这种便利的通信手段有时会潜在着不良因素——垃圾邮件,它可能不仅给人们的生活与工作带来困扰,也可能威胁着每个人的计算机系统或个人隐私的安全问题。因此寻找行之有效的垃圾邮件过滤方法已经成为研究者们感兴趣的一门研究课题。朴素贝叶斯分类算法是研究者们经常使用的一种用于过滤垃圾邮件的办法。它也是数据挖掘领域中简单易懂的一种分类算法,然而这种算法大多数情况下处理的是离散型数据。因此为了让朴素贝叶斯算法能够灵活地应对连续型数据,在分类的过程中往往需要对数据进行离散化处理。而模糊数学方面的相关理论恰恰可以用来解决这个问题。因此可以将二者结合来生成一个模糊贝叶斯混合模型,从而使整个分类模型去应对更多不同类型的数据,以此提高原有算法的分类性能。本课题研究了模糊贝叶斯混合模型在邮件系统中的应用,期间主要完成了以下工作:1.了解朴素贝叶斯算法、模糊数学以及三角模糊数的基本概念,收集相关理论知识和有关研究成果。找出朴素贝叶斯算法存在的缺陷,从而解决朴素贝叶斯为什么能够与模糊数学进行混合的问题。2.创建模糊贝叶斯混合分类模型的过程中,在去模糊化阶段引用了一种新的去模糊化方法“内心法”,该方法利用了三角形的内心来将三角模糊数转换成精确值。3.了解邮件系统的逻辑和功能结构,并通过查阅资料熟悉邮件传输的过程。4.收集邮件数据样本,并分别构造出用于分类的训练集和测试集。然后将混合分类模型应用于系统的邮件分类模块中,通过测试分析来总结混合模型性能。本文详细阐述了以上工作内容,同时总结、反思了研究中目前仍然存在的问题和需要改进的地方。
[Abstract]:In recent years, with the rapid development of network technology, more and more people choose to communicate with others through E-mail, a simple and fast way of communication. However, this convenient means of communication is sometimes potentially a bad factor-spam. It may not only bring troubles to people's life and work, but also threaten everyone's computer system or personal privacy security problems. Therefore, the search for effective spam filtering methods has become an interesting research topic. Naive Bayes classification algorithm is often used by researchers to filter spam. It is also a simple and easy to understand classification algorithm in the field of data mining, but in most cases, this algorithm deals with discrete data. Therefore, in order to make naive Bayesian algorithm flexible to deal with continuous data, it is often necessary to discretize the data in the process of classification. The theory of fuzzy mathematics can be used to solve this problem. So we can combine the two to generate a fuzzy Bayesian mixed model, so that the whole classification model can deal with more different types of data, so as to improve the classification performance of the original algorithm. In this paper, the application of fuzzy Bayesian hybrid model in mail system is studied, and the following work is accomplished: 1. The basic concepts of naive Bayes algorithm, fuzzy mathematics and triangular fuzzy numbers are understood, and relevant theoretical knowledge and related research results are collected. Find out the defects of naive Bayes algorithm, and solve the problem of why naive Bayes can be mixed with fuzzy mathematics. In the process of creating a fuzzy Bayesian mixed classification model, a new de-fuzzification method, "inner method", is introduced in the de-fuzzification stage. This method uses the heart of a triangle to convert the triangular fuzzy number into an exact value .3. Understand the logic and functional structure of the mail system, and familiarize yourself with the mail transfer process. Collect mail data samples and construct training set and test set for classification respectively. Then the hybrid classification model is applied to the mail classification module of the system, and the performance of the hybrid model is summarized by test and analysis. In this paper, the above work is described in detail, at the same time, the problems still existing in the research and the areas that need to be improved are reviewed.
【学位授予单位】：中国地质大学(北京)
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP393.098

【相似文献】