适用于大规模信息网络的语义社区发现方法
发布时间：2019-04-18 22:29
【摘要】：对节点带有内容的信息网络进行语义社区发现是新的研究方向。融合节点内容增加了算法的复杂度。提出了一种在线性时间内进行语义社区发现的标签传播算法,用LDA(latent Dirichlet allocation)主题模型表示节点内容,以节点内容相似度和传播影响力的乘性模型作为标签传播的策略,在归一化过程中,自然融合节点内容和网络结构信息,标签迭代过程中,采用节点与绝大部分邻居节点内容不相同才进行更新的策略,保证算法的运行效率。通过在不同规模的12个真实数据集上进行实验,以模块度和纯度作为度量标准,验证了算法在语义社区发现上的有效性和可行性。
[Abstract]:Semantic community discovery of node information networks with content is a new research direction. The fusion node content increases the complexity of the algorithm. In this paper, a label propagation algorithm for semantic community discovery in linear time is proposed. The node content is represented by LDA (latent Dirichlet allocation) topic model, and the multiplicative model of node content similarity and propagation influence is used as the strategy of tag propagation. In the process of normalization, the content of nodes and the information of network structure are merged naturally. In the process of label iteration, the updating strategy is adopted to ensure the running efficiency of the algorithm by adopting the strategy that the contents of nodes and most of the neighboring nodes are not the same. The effectiveness and feasibility of the algorithm in semantic community discovery are verified by experiments on 12 real data sets of different scales, taking modularity and purity as metrics.
【作者单位】： 北京联合大学商务学院;中国人民大学信息学院;北京交通大学计算机与信息技术学院;
【基金】：国家自然科学基金Nos.71572015,71271209 北京联合大学新起点项目No.Zk10201506~~
【分类号】：TP301.6
，
本文编号：2460345
[Abstract]:Semantic community discovery of node information networks with content is a new research direction. The fusion node content increases the complexity of the algorithm. In this paper, a label propagation algorithm for semantic community discovery in linear time is proposed. The node content is represented by LDA (latent Dirichlet allocation) topic model, and the multiplicative model of node content similarity and propagation influence is used as the strategy of tag propagation. In the process of normalization, the content of nodes and the information of network structure are merged naturally. In the process of label iteration, the updating strategy is adopted to ensure the running efficiency of the algorithm by adopting the strategy that the contents of nodes and most of the neighboring nodes are not the same. The effectiveness and feasibility of the algorithm in semantic community discovery are verified by experiments on 12 real data sets of different scales, taking modularity and purity as metrics.
【作者单位】： 北京联合大学商务学院;中国人民大学信息学院;北京交通大学计算机与信息技术学院;
【基金】：国家自然科学基金Nos.71572015,71271209 北京联合大学新起点项目No.Zk10201506~~
【分类号】：TP301.6
，
本文编号：2460345
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2460345.html
