【摘要】：隨著互聯(lián)網(wǎng)的飛速發(fā)展,如何有效地來對網(wǎng)絡流量和用戶行為進行監(jiān)管,構建一個文明健康、可信穩(wěn)定的網(wǎng)絡空間,漸漸引起了研究者們的注意。因此,如何對網(wǎng)絡中不同的人(即端目標)進行表征與識別開始成為當前研究者們關注的一個焦點。近年來研究較多的是如何利用流行為特征對網(wǎng)絡流進行分類,而將其應用于網(wǎng)絡端目標的表征與識別的研究則相對較少。針對上述網(wǎng)絡端目標表征與識別的研究現(xiàn)狀,本文提出基于服務類型劃分的分析方法,首先根據(jù)不同的服務類型對流量進行分類,并應用于網(wǎng)絡流的行為特征的提取和選擇,得到網(wǎng)絡端目標的表征,隨后引入機器學習和社團發(fā)現(xiàn)算法,最終完成網(wǎng)絡端目標的識別,并取得了不錯的效果。主要工作如下:(1)針對個體端目標的識別,即識別一個特定的用戶行為是由哪個端目標產(chǎn)生的,本文引入了基于機器學習的分類方法。首先將用戶的流量梳理到作者劃分的24種服務類型之下,用于構建端目標的流量矩陣,接著就是對原始的數(shù)據(jù)包處理得到分析所需的相關流行為特征,經(jīng)過特征選擇之后最后得到用于表征一個端目標的特征參數(shù)集,如此一天的流量數(shù)據(jù)便可以轉(zhuǎn)化為表征該端目標行為的一個樣本。采集了足夠多的樣本數(shù)據(jù)之后,便得到了機器學習所需的樣本數(shù)據(jù),經(jīng)過對樣本數(shù)據(jù)的手工標記之后,本文采用機器學習中的C4.5決策樹算法將樣本數(shù)據(jù)用于訓練和測試,最終取得了不錯的識別效果。(2)針對個體端目標之間的行為相似性,即發(fā)現(xiàn)網(wǎng)絡中潛在的社團群體,本文提出了基于流行為特征分析的社團發(fā)現(xiàn)算法來進行分析。由于需要衡量端目標之間的行為相似性,作者分別使用Dice相似度計算流行為特征的相似度,余弦相似度計算服務類型的相似度,構建相似度矩陣。最后利用社團發(fā)現(xiàn)算法分別得出基于流行為特征和服務類型的社團結構劃分,綜合兩者的結果得到最終的社團劃分結果。
[Abstract]:With the rapid development of the Internet, how to regulate the network traffic and user behavior effectively and build a civilized, healthy, credible and stable network space has gradually attracted the attention of researchers. Therefore, how to characterize and identify different people in the network has become a focus of attention. In recent years, much research has been done on how to classify network flows by using popular features, but relatively few studies have been made on their application to the characterization and recognition of network end targets. In view of the research status of target representation and recognition on the network side, this paper proposes an analysis method based on the classification of service types. Firstly, traffic is classified according to different service types, and it is applied to the extraction and selection of behavior characteristics of network flows. Then the machine learning and community discovery algorithms are introduced to realize the recognition of the target in the network, and good results are obtained. The main work is as follows: (1) for the recognition of individual target, that is, to identify which end target a particular user behavior is generated by, this paper introduces a classification method based on machine learning. First of all, the user traffic is combed under the 24 kinds of service types divided by the author, which is used to construct the traffic matrix of the end target, and then it is characterized by the related popularity needed for the analysis of the original data packet processing. After feature selection, a feature parameter set is obtained to represent an end target, so that the traffic data of a day can be transformed into a sample to represent the behavior of the end target. After collecting enough sample data, the sample data needed for machine learning is obtained. After manual marking of the sample data, this paper uses C4.5 decision tree algorithm in machine learning to train and test the sample data. Finally, a good recognition effect is achieved. (2) aiming at the behavior similarity between individual targets, that is, to find the potential community groups in the network, this paper proposes a community discovery algorithm based on popular feature analysis to analyze the behavior. Due to the need to measure the behavioral similarity between the end targets, the author uses Dice similarity to calculate the similarity of popular features and cosine similarity to calculate the similarity of service types, and constructs a similarity matrix. Finally, the community structure partition based on the popular feature and service type is obtained by using the community discovery algorithm, and the final community partition result is obtained by synthesizing the two results.
【學位授予單位】：電子科技大學
【學位級別】：碩士
【學位授予年份】：2017
【分類號】：TP393.06

【參考文獻】

相關期刊論文 前3條

1 李喬;何慧;方濱興;張宏莉;王雅山;;基于信任的網(wǎng)絡群體異常行為發(fā)現(xiàn)[J];計算機學報;2014年01期

2 劉興彬;楊建華;謝高崗;胡s，

本文編號：2359073