刻画多种潜在关系的泊松-伽马主题模型  

Poisson-gamma topic model of describing multiple underlying relationships

作　　者：吴艳丽 刘淑薇 何东晓[1] 王晓宝 金弟 WU Yanli;LIU Shuwei;HE Dongxiao;WANG Xiaobao;JIN Di(College of Intelligence and Computing,Tianjin University,Tianjin 300350,China)

机构地区：[1]天津大学智能与计算学部,天津300350

出　　处：《山东大学学报（工学版）》2023年第2期51-60,共10页Journal of Shandong University（Engineering Science）

基　　金：国家自然科学基金面上项目(61876128)。

摘　　要：为探索节点间链接结构的多种潜在关系并对其进行语义解释,提出一个刻画多种潜在关系的泊松-伽马主题模型,刻画不同潜在关系下节点内容与链接结构(边)的生成过程,利用全期望定律来聚合所有潜在关系中的内容信息与拓扑信息。对于模型推断,进一步提出一种封闭式的吉布斯采样算法。在8个真实数据集上与8种代表性社团发现方法进行比较,并对所有潜在关系中的链接结构进行可视化和案例分析。试验结果表明,本研究方法优于8种代表性的社团发现方法,能够在多种潜在关系中探索节点间链接结构的有效性,还能够利用节点内容来解释链接关系中的语义信息。In order to explore various potential relationships of link structures between nodes and interpret them semantically,a Poissongamma topic model that described multiple potential relationships was proposed.This model described the generation process of node contents and link structures(edges)under different potential relationships,and the law of total expectation was used to aggregate the contents and topology in all potential relationships.For the model inference,a closed Gibbs sampling algorithm was proposed.This study compared eight representative community discovery methods on eight real datasets,and visualized and analyzed the link structures in all potential relationships.The experimental results showed that the proposed research method was superior to eight representative community discovery methods,which could not only explore the effectiveness of link structures among nodes in a variety of potential relationships,but also used node contents to explain the semantic information in link relationships.

关 键 词：社交网络 社团发现 概率图模型 主题模型 语义 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]