讲座题目：Detecting communities in the attributed network through regularized clustering and its application
主 讲 人：杨虎
杨虎，中央财经大学信息学院副教授，硕士生导师，研究方向为大数据分析与统计计算、社会计算、复杂网络等。承担了国家社科基金后期资助、国家自然科学基金等项目，发表论文20余篇，获得国家发明专利授权5项，研究成果已发表在Statistics in Medicine、Information Processing & Management、《经济社会体制比较》等国内外知名学术期刊。兼任JoSC副主编，中国人工智能学会社会计算与社会智能专业委员会委员，及IPM、IJIM、IJPR、系统科学理论与实践等期刊审稿人。
How to exploit the heterogeneous information in the attributed networks to improve the performance of community detection is a hot research topic. In recent years related network embedding methods have been extended to implement community detection, especially state-of-art deep learning methods (such as VGAE) which translate attributed network to vectorized data by capturing heterogeneity information in the network and make traditional machine learning methods can deal with the attributed network for community detection. However, both problems of selecting the dimension of representations in network embedding and determining the number of communities have not yet been addressed. Thus, this study develops a learning framework for community detection and settles the challenging issues of embedding dimension selection and the determination of the number of unknown communities. To the best of our knowledge, this study is the first to combine these techniques with the following advantages: (1) It settles issues of dimension selection and the number of communities determined through optimizing regularized k-means automatically. (2) Both the computational algorithm and the statistical theorems are given to illustrate why the new method can overcome the influence of redundant embedding, and determine the unknown number of communities. (3) Application for community detection of two benchmark data and syndicated investment networks in China supports our theoretical analysis and shows the advantage of the new method.