为有效实现海量数据的非线性聚类,提出基于GraphLab的分布式流式近邻传播算法——GStrAP(GraphLab based stream affinity propagation)。该算法将数据抽象为有向无环图模型,采用"Gather-Apply-Scatter"的模式完成数据同步和...为有效实现海量数据的非线性聚类,提出基于GraphLab的分布式流式近邻传播算法——GStrAP(GraphLab based stream affinity propagation)。该算法将数据抽象为有向无环图模型,采用"Gather-Apply-Scatter"的模式完成数据同步和算法迭代。在人工合成流形数据3D Clusters、Aggregation、Flame和Pathbased数据集上分别采用不同数据规模以及与传统K-means的聚类性能做对比,实验表明:基于GraphLab的近邻传播算法对数据规模具有良好的拓展性,在保持算法聚类效果的同时,有效降低时间复杂度。展开更多
随着网络规模的不断扩大,经典的复杂网络重叠社识别算法已不能高效处理现有的大规模网络图数据.本文在GraphLab并行计算模型上提出了基于重要节点扩展的重叠社区识别算法DOCVN(Detecting the Overlapping Community algorithm based on ...随着网络规模的不断扩大,经典的复杂网络重叠社识别算法已不能高效处理现有的大规模网络图数据.本文在GraphLab并行计算模型上提出了基于重要节点扩展的重叠社区识别算法DOCVN(Detecting the Overlapping Community algorithm based on Vital Node Expanding in GraphLab).算法选取网络中PageRank值大的节点作为重要节点,计算其他节点归属于重要节点的节点归属度,并以重要节点为中心形成核心社区及扩展社区,最后根据重要节点间的连接紧密度合并核心社区及扩展社区,并计算出每个节点在所属社区里的节点重要度,实现了大规模网络的重叠社区识别.实验表明该算法与PD(Propinquity Dynamics)等现有并行算法相比更能有效地识别大规模网络的重叠社区结构.展开更多
This research examines industry-based dissertation research in a doctoralcomputing program through the lens of machine learning algorithms todetermine if natural language processing-based categorization on abstractsal...This research examines industry-based dissertation research in a doctoralcomputing program through the lens of machine learning algorithms todetermine if natural language processing-based categorization on abstractsalone is adequate for classification. This research categorizes dissertationby both their abstracts and by their full-text using the GraphLabCreate library from Apple’s Turi to identify if abstract analysis is anadequate measure of content categorization, which we found was not. Wealso compare the dissertation categorizations using IBM’s Watson Discoverydeep machine learning tool. Our research provides perspectiveson the practicality of the manual classification of technical documents;and, it provides insights into the: (1) categories of academic work createdby experienced fulltime working professionals in a Computing doctoralprogram, (2) viability and performance of automated categorization of theabstract analysis against the fulltext dissertation analysis, and (3) natuallanguage processing versus human manual text classification abstraction.展开更多
文摘随着网络规模的不断扩大,经典的复杂网络重叠社识别算法已不能高效处理现有的大规模网络图数据.本文在GraphLab并行计算模型上提出了基于重要节点扩展的重叠社区识别算法DOCVN(Detecting the Overlapping Community algorithm based on Vital Node Expanding in GraphLab).算法选取网络中PageRank值大的节点作为重要节点,计算其他节点归属于重要节点的节点归属度,并以重要节点为中心形成核心社区及扩展社区,最后根据重要节点间的连接紧密度合并核心社区及扩展社区,并计算出每个节点在所属社区里的节点重要度,实现了大规模网络的重叠社区识别.实验表明该算法与PD(Propinquity Dynamics)等现有并行算法相比更能有效地识别大规模网络的重叠社区结构.
文摘This research examines industry-based dissertation research in a doctoralcomputing program through the lens of machine learning algorithms todetermine if natural language processing-based categorization on abstractsalone is adequate for classification. This research categorizes dissertationby both their abstracts and by their full-text using the GraphLabCreate library from Apple’s Turi to identify if abstract analysis is anadequate measure of content categorization, which we found was not. Wealso compare the dissertation categorizations using IBM’s Watson Discoverydeep machine learning tool. Our research provides perspectiveson the practicality of the manual classification of technical documents;and, it provides insights into the: (1) categories of academic work createdby experienced fulltime working professionals in a Computing doctoralprogram, (2) viability and performance of automated categorization of theabstract analysis against the fulltext dissertation analysis, and (3) natuallanguage processing versus human manual text classification abstraction.