III: Small: Collaborative Research: A General Feature Learning Framework for Dynamic Attributed Networks

III：小：协作研究：动态属性网络的通用特征学习框架

• 批准号: 1715940
• 负责人: Jiliang Tang
• 金额: $ 24.79万
• 依托单位: Michigan State University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2017
• 资助国家: 美国
• 起止时间: 2017-08-01 至 2020-07-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1715940&HistoricalAwards=false
• 关键词: III; Small; Collaborative; Research; General

Attributed networks are those networks which are associated with a rich set of attributes. For example, in online social networks, users post messages related to what they are experiencing, which can be represented as a series of word attributes; in health care systems, providers are networked with each other given their shared patients, and each provider has profile information and may submit insurance claims as attribute information. Feature learning aims at seeking effective representations of data instances in preparing the attributed networks for various data mining tasks. Feature learning algorithms, including feature extraction and feature selection, have been intensively studied in the literature. While most existing studies focused on static, pure and shallow networks, this project aims to develop novel feature learning algorithms for dynamic attributed networks. The output of the project will be a series of feature learning algorithms, including shallow and deep network embedding, and feature selection, specifically designed for dynamic attributed networks. The developed algorithms, as well as their corresponding theoretical understandings, are expected to significantly advance data-driven social computing and health informatics. The goal of this project is to develop a novel feature learning framework for dynamic attributed networks, which consists of network embedding and corresponding deep architectures, as well as feature selection algorithms. The feature learning framework is feasible to effectively and efficiently address data challenges raised by dynamic attributed networks from various aspects. Specifically, this project aims to achieve the research goal through three primary research objectives: (1) performing dynamic network embedding under challenging scenarios, including the limited label information, heterogeneous feature spaces, and scalability of the data; (2) designing deep architectures for dynamic network embedding on various types of attributed networks; and (3) developing feature selection methods, by modeling dynamic attributed networks with link weights and cross-media links, to further enable interpretability in network analytics. In addition, this project will incorporate the research problems in a new curriculum, and it will also allow the PIs to continue the ongoing efforts to provide research opportunities to undergraduate and underrepresented students.

属性网络是那些与丰富的属性集相关联的网络。例如，在在线社交网络中，用户发布与他们所经历的事情相关的消息，这些消息可以表示为一系列单词属性；在医疗保健系统中，提供者彼此联网，给定他们共享的患者，每个提供者都有个人资料信息，并可以将保险索赔作为属性信息提交。特征学习的目的是在为各种数据挖掘任务准备属性网络时，寻求数据实例的有效表示。特征学习算法，包括特征提取和特征选择，已经在文献中得到了广泛的研究。虽然现有的研究大多集中在静态、纯和浅层网络上，但本项目的目标是为动态属性网络开发新的特征学习算法。该项目的输出将是一系列特征学习算法，包括浅层和深层网络嵌入，以及专门为动态属性网络设计的特征选择。所开发的算法以及它们相应的理论理解，预计将显著推动数据驱动的社会计算和健康信息学。该项目的目标是开发一种新的动态属性网络的特征学习框架，该框架包括网络嵌入和相应的深层结构，以及特征选择算法。该特征学习框架能够从多个方面有效、高效地应对动态属性网络带来的数据挑战。具体地说，本项目旨在通过三个主要研究目标实现研究目标：(1)在具有挑战性的场景下执行动态网络嵌入，包括有限的标签信息、异质特征空间和数据的可扩展性；(2)为动态网络嵌入到各种类型的属性网络设计深层体系结构；(3)开发特征选择方法，通过对带有链路权重和跨媒体链接的动态属性网络建模，进一步实现网络分析中的可解释性。此外，这个项目将把研究问题纳入新的课程，并将允许私人投资经理继续努力，为本科生和代表性不足的学生提供研究机会。