基于P2P视频共享网络的社区化兴趣网络研究
文献类型：学位论文
| 作者 | 牛尔力 |
| 学位类别 | 博士 |
| 答辩日期 | 2008-06-02 |
| 授予单位 | 中国科学院声学研究所 |
| 授予地点 | 声学研究所 |
| 关键词 | 对等网络 社区结构 分布式社区发现 视频共享 |
| 其他题名 | Studies on Interest-Based Network on Video Sharing Peer-to-Peer Network |
| 学位专业 | 信号与信息处理 |
| 中文摘要 | 近年来，随着网络技术的发展，视频共享得到了广泛的应用。视频共享网络包括基于客户端/服务器方式（C/S方式）和基于对等方式（P2P方式）两种架构的应用。在基于C/S架构的应用（例如Youtube等）中，服务器掌握所有视频资源的信息，能够为用户提供丰富的服务，然而其扩展性和健壮性较差，维护成本比较高。基于P2P架构的应用（例如Gnutella、eMule等）具有良好的扩展性和健壮性，易于维护，因而受到广泛关注。但是由于资源分布在各个节点上，网络拓扑的组织也是随机的，每个节点只知道局部的资源信息，因此无法像C/S架构那样提供丰富的服务，如果能够根据节点共享的资源形成网络拓扑，节点和资源的分布比较有规律，则比较容易查找和利用资源，从而能有效地开展多种业务。 P2P网络中节点在共享资源上具有异构性。基于共享资源表示节点兴趣，连接具有共同兴趣的节点以形成基于兴趣的P2P网络（以下简称兴趣网络），并基于兴趣网络改进资源发现是P2P中的研究热点之一。然而，这些研究多集中于基于兴趣网络改进资源发现的路由算法，未能充分考虑网络的拓扑结构和社区结构对算法的影响，也没有利用兴趣网络开展多种业务。针对这一现状，本文深入研究兴趣网络的应用、测量、生成以及社区结构发现。 本文的主要贡献和创新工作如下： 1、提出一种基于P2P视频共享网络的社区化兴趣网络结构，有助于实现P2P视频共享网络中基于社区化兴趣网络的搜索服务和视频推荐服务。包括基于节点在共享文件上的异构性，在非结构化P2P视频共享网络上生成具有社区结构的兴趣网络（简称社区化兴趣网络）；基于社区化兴趣网络采用基于用户的协同过滤和基于项目的协同过滤技术，为用户提供视频推荐服务；基于社区化兴趣网络采用随机漫步和洪泛相结合的方式改进搜索机制，同时避免针对热门资源和无关社区的洪泛搜索。 2、针对社区化兴趣网络生成中如何度量节点相似度的问题，提出一种基于TFIDF模型的社区化兴趣网络及其测量方法。该方法根据全局节点信息计算基于TFIDF（Term Frequency，词频，Inverse Document Frequency，逆文本频率指数）模型的节点相似度，基于相似度构建社区化兴趣网络，并测量了该网络的拓扑结构和社区结构。试验表明，在基于TFIDF模型的社区化兴趣网络中，节点度分布接近幂率分布，并且具有较强的社区结构，更适合实际应用。 3、针对如何基于局部信息生成社区化兴趣网络和减小负载的问题，提出社区化兴趣网络生成机制，包括分布式IDF计算方法、元数据交互机制和友邻节点维护机制。分布式IDF计算方法利用局部信息生成样本集，基于样本集估计单词的IDF值；哈希表和数据压缩相结合的元数据交互机制减小了网络负载；k 步随机漫步方式的探测机制、自适应探测机制和基于TFIDF的特征选择方法有效的减少探测信令的数目和大小，从而减小维护友邻节点产生的网络负载。试验表明，这些机制有效的减小网络中的负载，形成的网络具有较强的社区结构。 4、针对如何基于局部信息发现社区化兴趣网络中社区结构的问题，提出一种分布式社区发现方法PDC（Power based Distributed Community Detecting Algorithm，简称PDC）。该方法根据局部信息计算节点Power值，基于节点的Power值选择每个社区的中心节点，并通过中心节点发送消息以发现社区结构。试验表明，该方法能有效的发现社区结构，并能表示节点和各个社区的相关性。 |
| 英文摘要 | With the rapid development of computer networks, video sharing has gained much attention in recent years. Some type of video sharing application adopts client/server model, where all the information of video is held by server. So it is able to provide many services easily. But it has some drawbacks, including complexity of network configuration and management, less robustness and worse scalability. Another type of video sharing application adopts peer-to-peer model, which has the advantages in robustness, scalability, and cost-efficiency, and can be implemented with zero configurations. But all the information of video is distributed in peers and each peer only has local information, so it provides simple and few services. If the video sharing peer-to-peer network is constructed basing on resource to generate interest-based network, it was easy to search and utilize resources. Currently many researches are about utilizing the interest-based network to improve search, but few of them measure characteristics of interest-based network and utilize it to provide new services. Reducing costs of constructing it is also important in its application. Following from the above observations, this dissertation propose some novel and effective approaches to address above problems. The main contributions and innovative works of the dissertation include: 1、An interest-based peer-to-peer network in video sharing system is proposed, which is helpful to improve search and to adopt collaborative filtering. The interest-based network is constructed on unstructured peer-to-peer network. User-based and item-based collaborative filtering methods are adopted to provide video recommender service. The improved searching method combines random walk with flooding, which avoid flooding searching for hot resource and in unrelated communities. 2、A method to construct and measure TFIDF-based network is proposed. This method constructs TFIDF-based network using data captured from Gnutella. Topology and community structure of the constructed network is measured. Experimental results show that TFIDF-based network is appropriate for application. Its node degree distribution is similar to power law and it has strong community structure. 3、A mechanism to construct interest-based network is proposed. It includes a distributed IDF calculating method, a metadata exchanging mechanism and a friend detecting mechanism. IDF is calculated using samples. The metadata exchanging method combines hash table with compression, which reduce network loads. The friend detecting mechanism adopts k random walk mechanism, adaptive method and TFIDF-based method to reduce message number and size. Experimental results show that these methods reduce load effectively and the constructed network present strong community structure. 4、A distributed community detecting method is proposed. This method defines node Power and selects community center node according to it, and each center node send message to detect community structure. Experimental results show that this method make peers known its community and its relation with communities |
| 语种 | 中文 |
| 公开日期 | 2011-05-07 |
| 页码 | 121 |
| 源URL | [http://159.226.59.140/handle/311008/324]  |
| 专题 | 声学研究所_声学所博硕士学位论文_1981-2009博硕士学位论文 |
| 推荐引用方式 GB/T 7714 | 牛尔力. 基于P2P视频共享网络的社区化兴趣网络研究[D]. 声学研究所. 中国科学院声学研究所. 2008. |
入库方式： OAI收割
来源：声学研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。