时间：2024年5月10日（星期五）14:00-15:00
地点：经管大楼A楼 四楼第二会议室报告厅
主题：基于自监督学习的网络缺失节点检测研究（Missing Node Detection on Complex Networks Based on Self-Supervised Learning）
主讲人：周立欣（MD传媒app下载）
介绍：周立欣，信管系副研究员，硕士生导师。研究方向包括应急管理、数据挖掘、图神经网络等，主持多项国家级和省部级科研项目，包括国家社科基金青年项目1项，博士后科学基金面上项目和上海市人民政府决策咨询研究基地项目各一项。以第一作者或通讯作者在Information Science、Knowledge-Based Systems、Technological Forecasting and Social Change等国内外权威期刊发表论文12篇（中科院TOP期刊4篇，JCR 1区9篇）。
Zhou Lixin, Associate Proffoser the Information Management Department, and Master Supervisor. His research directions include emergency management, data mining, graph neural networks, etc. He has presided over several national and provincial-level scientific research projects, including one National Social Science Fund, Postdoctoral Science Foundation general project, and Research based project of The Development Research Center of Shanghai Municipal People's Government. As the first author or corresponding author, he has published 12 papers in authoritative journals at home and abroad, including 4 papers in Chinese Academy of Sciences TOP journals and 9 papers in JCR Q1 journals.
摘要：网络广泛存在于许多现实世界的系统中，包括社交网络、生物网络和交通网络。近年来，在这些网络上进行的数据挖掘和分析已经变得非常流行。然而，由于获取图结构化数据的成本高或者无法获得访问权限，我们获取的数据通常是不完整的，包含许多缺失的节点和边。缺失的节点和边可能导致图的结构和功能发生变化，进而影响节点分类、社区检测、推荐系统和其他下游任务。针对网络存在缺失节点的问题，结合图学习、自监督学习和对比学习等深度学习理论和模型，提出基于图卷积的自监督缺失节点检测模型和引入图对比学习的缺失节点检测及校验模型。通过在引文网络和社交网络等多个领域的数据集的实验分析，验证了模型和方法的有效性。
Networks are widely present in many real-world systems, including social networks, biological networks, and transportation networks. In recent years, data mining and analysis on these networks have become very popular. However, due to the high cost of acquiring graph-structured data or the lack of access permissions, the data we obtain is often incomplete, containing many missing nodes and edges. These missing nodes and edges can lead to changes in the structure and function of the graph, which in turn affects node classification, community detection, recommendation systems, and other downstream tasks. To address the issue of missing nodes in networks, this text proposes a self-supervised missing node detection model based on graph convolution and a missing node detection and verification model that introduces graph contrastive learning, combining deep learning theories and models such as graph learning, self-supervised learning, and contrastive learning. The effectiveness of the models and methods has been verified through experimental analysis on datasets from multiple domains, including citation networks and social networks.
