基于动态主题模型融合多维数据的微博社区发现算法

发布时间：2018-06-23 23:00

本文选题：新浪微博 + 文本挖掘　；参考：《软件学报》2017年02期

【摘要】：随着微博用户的不断增加,微博网络已成为用户进行信息交流的平台.针对由于博文长度受限,传统的社区发现算法无法有效解决微博网络的稀疏性等问题,提出了DC-DTM(discovery community by dynamic topic model)算法.DC-DTM算法首先将微博网络映射为有向加权网络,网络中边的方向反映节点之间的关注关系,利用所提出的DTM(dynamic topic model)计算出节点之间的语义相似度,并将其作为节点间连边的权重.DTM是一种微博主题模型.该模型不仅能够挖掘博客的主题分布,而且能够计算出某一主题中用户的影响力大小.其次,利用所提出的复杂度较低的标签传播算法WLPA(weighted lebel propagation)进行微博网络的社区发现.该算法的初始化阶段将影响力大的用户节点作为初始节点,标签按照节点的影响力从大到小进行传播,避免了传统标签传播算法逆流现象的发生,提高了标签传播算法的稳定性.真实数据上的实验结果表明,DTM模型能够很好地对微博进行主题挖掘,DC-DTM算法能够有效地挖掘出微博网络的社区.
[Abstract]:With the increasing of Weibo users, Weibo network has become a platform for users to exchange information. Because of the limited length of blog posts, the traditional community discovery algorithm can not effectively solve the problem of Weibo network sparsity and so on. A DC-DTM (discovery community by dynamic topic model) algorithm .DC-DTM algorithm is proposed to map the Weibo network to a directed weighted network. The direction of edges in the network reflects the relationship of concern between nodes, and the semantic similarity between nodes is calculated by using the proposed DTM (dynamic topic model), which is regarded as the weight of the connected edges between nodes. DTM is a kind of Weibo topic model. The model not only can mine the distribution of blog topics, but also can calculate the influence of users in a certain topic. Secondly, the proposed tag propagation algorithm WLPA (weighted lebel propagation) is used for community discovery in Weibo networks. In the initialization phase of the algorithm, the influential user nodes are taken as the initial nodes, and the labels are propagated according to the influence of the nodes from large to small, thus avoiding the reverse flow phenomenon of the traditional label propagation algorithm. The stability of label propagation algorithm is improved. The experimental results on real data show that the DTM model can effectively mine the community of Weibo network by using DC-DTM algorithm.
【作者单位】：东北大学计算机科学与工程学院;东北大学软件学院;
【分类号】：TP181

【相似文献】