超级节点网络中并行查询和合并机制研究

针对超级网络中超级节点可能会导致网络瓶颈、检索结果重复问题,提出一种并行查询和合并机制。首先提出超级节点选择查询节点的算法,以减少超级节点的存储和计算负担。然后提出选择下载节点的合并算法,以得到高质量、新颖和更能有效获取的检索结果。实验结果表明了该机制的有效性。     

超级节点网络的构建与搜索机制研究

超级节点网络是提高P2P网络搜索效率的有效方式之一.提出基于在线聚类的超级节点网络构建和路由方法.采用在线聚类算法,将新加入的节点按照语义相关性,动态加入或创建相应的超级节点,构建超级节点网络.算法在保证每个超级节点所连接的普通节点之间具有很强的语义相关性的基础上,当超级节点出现负载过重时,采用动态调整超级节点选择策略.该策略解决了随着网络规模的扩大,超级节点因负载过重而出现网络瓶颈的问题.在此基础上,提出一种优化搜索机制,提高了搜索效率和检索结果.     

数据网格环境下的多媒体资源检索

提出一种数据网格环境下的多媒体资源检索方法。通过设计分层结构的虚拟资源空间管理非结构化异构资源,当网格查询节点收到检索请求时,根据虚拟集和元数据等信息进行信息过滤,得到候选集及相应的结构化特征和语义标注,执行节点对候选集进行相似度求精和排序运算,并将排序结果返回查询节点。仿真结果表明,该方法具有较好的时间性能和较高的查准率。     

移动P2P网络超级节点选取机制研究

针对半分布式结构的移动P2P网络超级节点选取问题,提出一种新颖的超级节点选取机制,全面分析影响超级节点选取的各种因素,按着其特点和类型不同将其分为效益型属性和成本型属性,并在此基础上建立了超级节点选取的带约束多目标优化模型.采用新的约束处理方法,利用免疫克隆算法对超级节点选取问题进行求解.实验结果表明,该机制可以有效降低超级节点的失效率,提高系统查询效率,算法收敛速度快,有较好的可扩展性.     

移动P2P网络中超级节点的选择

针对移动P2P网络中节点移动性强、网络拓扑结构变化剧烈、导致系统的低效和不可靠问题,提出一种移动P2P网络中的超级节点选择算法,该算法能够选择性能好、在线时间长的节点作为超级节点,并且采用候补超级节点技术,有效改善移动P2P网络系统的效率。仿真实验结果表明,该算法可以有效降低超级节点的失效率和缩短查询延迟。     

一种基于超级节点的半分布式P2P系统改进策略

基于半分布式结构的P2P系统,兼有集中式和分布式结构的优点,具有良好的管理性和扩展性,市场上应用非常广泛.但由于网络中超级节点易受攻击,网络不稳定,而且该结构采用Gnutella网络的查询信息泛洪机制,造成了严重的网络带宽负担.提出了一种将数个超级节点组合在一起构成超级节点组的改进的半分布式P2P网络模型,平衡了网络负载,并且可以在很小的跳数内访问到网络中的绝大多数节点,不会产生过多的消息,节省了带宽,避免网络拥塞.     

基于P2P的路由查询机制

针对现有P2P检索的路由盲目性问题,提出能够适应网络可扩展性的路由查询机制。在检索过程中根据各节点的响应顺序,将每条查询路径上的回复节点信息分别保存在与它相邻的2个回复节点的路由信息表中,据此为以后的检索提供路由。在没有路由信息可用时,选择原始邻居节点进行路由,以利用原始拓扑结构的特点。实验结果表明,该机制能有效减少消息转发数量,提高信息检索的成功率。     

一种新的随机游走的异构网络节点排序算法研究

异构网络中往往包含多种不同类型的节点。为了满足用户对不同类型节点的检索需求,需要对不同类型的节点进行排序。同时鉴于用户对不同类型节点的检索需求不同, 提出了一种基于随机游走的异构网络节点排序算法,以提高检索结果的性能。首先,对包含用户、图片和文本内容的异构网络进行了形式化描述。其次,对异构网络中节点之间的相似性进行了定义,并提出了一种包含不同类型邻居节点的相似性计算方法。接下来,基于随机游走思想提出了节点重要性的计算算法,并分析了随机游走模型中偏好向量的选取方法。最后采用大量真实数据进行了实验验证,结果证明了提出的方法在处理信息检索请求时的有效性。     

混合式P2P系统中基于缓存的多项查询方法研究

首先从混合式P2P网络拓扑结构出发,结合DHT思想,提出了基于DHT的层次化P2P网络模型.其次根据在文档集巨大的情况下,用户提交的查询不可能"面面俱到",实际用来回答查询的文档仅仅是文档集中很小的一部分这一思想,在层次化P2P模型的超级节点中建立了分布式缓存,运用分布式索引与缓存技术,提出一种新的方法来解决多项查询问题.即由多项查询中的某个关键字key,根据hash函数定位到负责该key的超级节点,查询该节点上的分布式索引得到缓存具体存储位置,最终将结果返回给用户,如若缓存中没有所要查询的内容,则广播该查询,同时根据系统中的历史广播查询信息来计算某个待选缓存项的利益值,利益最大的待选项加入缓存.一般针对多项查询的泛洪算法往往会造成巨大的网络信息量,提出的方法牺牲了超级节点上一小部分的存储力,缓解了多项查询造成的网络拥挤现象.同时,基于DHT的层次化P2P模型也具有很好的稳定性,不会因为大量节点的动态加入或者退出而无法进行多项查询.     

一种高效的P2P环境中的窗口查询算法

随着多媒体以及P2P网络的发展,针对高维数据基于属性的窗口查询已经成为一个重要研究课题.提出了一种在超级节点P2P网络中有效解决高维数据的窗口查询算法,在每个单独的网络节点上,数据通过一种降维算法映射到一维空间,在超级节点上,构造数据的统计信息表以及构造网络查询树,算法在每次查询时,按照查询树的规则来访问整个网络,并利用统计信息剪枝网络中的节点查询,避免网络的泛洪.实验中使用了不同的数据集来评测算法的查询效率,结果表明该算法具有很高的查询效率.     

P2P搜索新技术：智能搜索技术

在P2P网络中正确高效地搜索信息十分重要,然而现有的搜索技术并不能很好地满足用户要求。文中提出了由历史记录、相似度和站点优先级3部分组成的新的信息搜索技术———智能搜索技术。在智能搜索技术中每个站点都各自记录以往搜索的历史记录,以此结合搜索要求计算出最可能拥有所需结果的站点,最后仅向这些站点发送搜索请求。通过在中间件上的测试,比较了不同信息搜索技术的性能,智能搜索技术能在使用较少搜索消息的情况下得到较高的搜索成功率。     

基于对抗式数据增强的深度文本检索重排序

在信息检索领域的排序任务中, 神经网络排序模型已经得到广泛使用. 神经网络排序模型对于数据的质量要求极高, 但是, 信息检索数据集通常含有较多噪音, 不能精确得到与查询不相关的文档. 为了训练一个高性能的神经网络排序模型, 获得高质量的负样本, 则至关重要. 借鉴现有方法doc2query的思想, 本文提出了深度、端到端的模型AQGM, 通过学习不匹配查询文档对, 生成与文档不相关、原始查询相似的对抗查询, 增加了查询的多样性,增强了负样本的质量. 本文利用真实样本和AQGM模型生成的样本, 训练基于BERT的深度排序模型, 实验表明,与基线模型BERT-base对比, 本文的方法在MSMARCO和TrecQA数据集上, MRR指标分别提升了0.3％和3.2％.     

基于多表达的第一阶段语义检索模型

当前,信息检索系统通常采用“检索+重排序”的多级流水线架构。基于稠密表示的检索模型已经被逐渐应用到第一阶段检索中,并展现出了相比传统的稀疏向量空间模型更好的性能。考虑到第一阶段检索所需的高效性,大多数情况下这些模型的基本架构都采用双编码器（bi-encoder）结构。对查询和文档进行独立的编码,分别得到一个稠密表示向量,然后基于获得的查询和文档表示使用简单的相似度函数计算查询-文档对的得分。然而,在编码文档的过程中查询是不可知的,而且文档相比查询而言通常包含更多的主题信息,因此这种简单的单表示模型可能会造成严重的文档信息丢失。为了解决这个问题,设计了一种新的语义检索方法 MDR(multi-representation dense retrieval),将文档编码成多个稠密向量表示。同时,该方法引入覆盖率（coverage）机制来保证多个向量之间的差异性,从而能够覆盖文档中不同主题的信息。为了评估模型性能,在MS MARCO数据集上进行了段落排序和文档排序任务,实验结果证明了MDR方法的有效性。     

Data-Sharing P2P Networks with Semantic Approximation Capabilities

The synergy between peer-to-peer systems and semantic Web technologies supports large-scale sharing of semantically rich data, usually represented through schemas such as RDF. Peers rarely share the same vocabulary, so the resulting heterogeneity of data representations introduces new challenges for the efficient and effective retrieval of relevant information. The authors leverage the presence of semantic approximations between peers' schemas to improve query routing by identifying the peers that best satisfy the user's requests, and to inform users of the relevance of the returned answers through a ranking mechanism that promotes the most semantically related results.     

Scalable Retrieval and Mining With Optimal Peer-to-Peer Configuration

We proposed to utilize the scalable peer-to-peer network to perform the content-based image retrieval and mining, i.e, P2P-CBIRM. The decentralized unstructured P2P model with certain overheads, i.e., peer clustering and update procedures, is adopted to compromise with the structured one while still reserving flexible routing control when peers join/leave or network fails. The peer CBIRM engine is designed to utilize multi-instance query with multi-feature types to effectively reduce network traffic while maintaining high retrieval accuracy. It helps to enhance the knowledge discovery and image data mining capability. The proposed P2P-CBIRM system provides the scalable retrieval and mining function that the query scope and retrieval accuracy can be adaptively and progressively controlled. To improve the query efficiency (recall-rate/query-scope), it effectively utilizes both: 1) forwarding query message (forward phase) to reduce the query scope and 2) transmitting retrieval results (backward phase) such that activated peers keep filtering high similarity images on the link-path toward the query peer. Experiments show that the query efficiency of the scalable retrieval approach is better than previous methods, i.e., firework query model and breadth-first search. It provides a scalable knowledge discovery platform for efficient image data mining applications. We also proposed to optimally configure the P2P-CBIRM system such that, under a certain number of online users, it would yield the highest recall rate. Simulations demonstrate that, with the optimal configuration, recall rates can be improved to 2.5 to 3 times larger while the network traffic of each peer is reduced to 30% of the original, under the same number of on-line users.     

ReMIS：面向再制造信息共享的P2P网络模型*

针对当前再制造生产的特点和实际需求,设计了一种面向再制造信息共享的P2P网络模型ReMIS。为适应再制造网络的高度动态性,模型采用基于自然簇和主题簇构成的两层混合结构,上层由主题对等体组成结构化的主题网络,下层由普通对等体组成P2P网络。ReMIS采用基于主题簇的资源分类和检索机制,能够确保在处理海量再制造资源信息的检索请求时具有较高的路由性能和查询性能,并能减少再制造生产中的词汇语义冲突;同时,发布/订阅机制的应用使再制造企业获取实时信息成为可能。相关性能分析表明,ReMIS网络是高效可行的,可以适应再     

Exploiting locality for scalable information retrieval in peer-to-peer networks

An important problem in unstructured peer-to-peer (P2P) networks is the efficient content-based retrieval of documents shared by other peers. However, existing searching mechanisms are not scaling well because they are either based on the idea of flooding the network with queries or because they require some form of global knowledge.We propose the Intelligent Search Mechanism (ISM) which is an efficient, scalable yet simple mechanism for improving the information retrieval problem in P2P systems. Our mechanism is efficient since it is bounded by the number of neighbors and scalable because no global knowledge is required to be maintained.ISM consists of four components: A Profiling Structure which logs queryhit messages coming from neighbors, a Query Similarity function which calculates the similarity queries to a new query, RelevanceRank which is an online neighbor ranking function and a Search Mechanism which forwards queries to selected neighbors.We deploy and compare ISM with a number of other distributed search techniques over static and dynamic environments. Our experiments are performed with real data over Peerware, our middleware simulation infrastructure which is deployed on 75 workstations. Our results indicate that ISM outperforms its competitors and that in some cases it manages to achieve 100% recall rate while using only half of the network resources required by its competitors. Further, its performance is also superior with respect to the total query response time and our algorithm exhibits a learning behavior as nodes acquire more knowledge. Finally ISM works well in dynamic network topologies and in environments with replicated data sources.     

双特征空间的实体排序学习

随着大规模知识图谱的出现以及企业高效管理领域知识图谱的需求,知识图谱中的自组织实体检索成为研究热点。给定知识图谱以及用户查询,实体检索的目标在于从给定的知识图谱中返回实体的排序列表。从匹配的角度来看,传统的实体检索模型大都将用户查询和实体统一映射到词的特征空间。这样做具有明显的缺点,例如,将同属于一个实体的两个词视为独立的。为此,该文提出将用户查询和实体同时映射到实体与词两个特征空间方法,称为双特征空间的排序学习。首先将实体抽象成若干个域。之后从词空间和实体空间两个维度分别抽取排序特征,最终应用于排序学习算法中。实验结果表明,在标准数据集上,双特征空间的实体排序学习模型性能显著优于当前先进的实体检索模型。     

Developing a multi-layer reference design retrieval technology for knowledge management in engineering design

Engineering design is a knowledge-intensive process that encompasses conceptual design, detailed design, engineering analysis, assembly design, process design, and performance evaluation. Each of these tasks involves various areas of knowledge and experience. The sharing of such knowledge and experience is critical to increasing the capacity for developing products and to increasing their quality. It is also critical to reducing the duration and cost of the development cycle. Accordingly, offering engineering designers various methods for retrieving engineering knowledge is one of the most important tasks in managing engineering knowledge.This study develops a multi-layer reference design retrieval technology for engineering knowledge management to provide engineering designers with easy access to relevant design and related knowledge. The tasks performed in this research include (i) designing a multi-layer reference design retrieval process, (ii) developing techniques associated with multi-layer reference design retrieval technology, and (iii) implementing a multi-layer reference design retrieval mechanism. The retrieval process contains three main phases—‘customer requirement-based reference design retrieval’, ‘functional requirement-based reference design retrieval’ and ‘functional feature-based reference design retrieval’. This technology involves (1) customer requirement-based reference design retrieval, which involves a structured query model for customer requirements, a case-based representation of designed entities, a customer requirement-based index structure for historical design cases, and customer requirement-based case searching, matching and ranking mechanisms, (2) functional requirement-based reference design retrieval, which includes a structured query model for functional requirements, a functional requirement-based index structure for historical design cases, and functional requirement-based case searching, matching and ranking mechanisms, and (3) functional feature-based reference design retrieval, which is a binary code-based representation for functional features, an ART1 neural network for functional feature-based case clustering and functional feature-based case ranking.