基于反问题求解的社交网络链路分析方法研究

项目名称： 基于反问题求解的社交网络链路分析方法研究

项目编号： No.61301274

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 无线电电子学、电信技术

项目作者： 费高雷

作者单位： 电子科技大学

项目金额： 24万元

中文摘要： 获取社交网络中的链路、节点重要度和节点种类等信息是社交网络管理的基础工作。基于反问题求解的社交网络链路分析可以避免传统链路挖掘方法需要较大代价才能获得社交网络链路等信息的缺陷，成为了目前国内外学术界和工业界共同关注的前沿科学问题。本项研究针对目前基于反问题求解的链路分析方法难以在实际中应用的问题，围绕社交网络链路分析反问题求解中的系统模型建立、提高稳定性、降低多解性和非线性求解方法四个问题展开研究。力求通过解决相应的理论问题，提高基于反问题求解的链路分析方法获取实际社交网络有用信息的能力。主要创新包括：提出基于非线性拟合和有监督机器学习的系统模型建立方法；提出基于阻尼最小二乘、约束最优化、多目标最优化的反问题求解方法提高稳定性；提出基于随机模拟的迭代初始值估计方法和约束迭代的反问题求解方法降低多解性；针对社交网络中噪声模型和系统模型的特点，提出相应的非线性反问题求解方法。

中文关键词： 社交网络；链路分析；多解性；稳定性；非线性

英文摘要： Obtaining the information such as links, node importance and node type in social network is a foundation work of social network management. Link analysis technique based on solving inverse problem is capable of avoiding the disvantage of link mining which usually requires spending high cost to obtain the information such as social network links, and became one of the most popular problem concerned by academia and industry in the world. Aiming at the problem that existing methods for link analysis technique based on solving inverse problem can not be applied to actual social network, this study implements research by focusing on the problems of system model construction, stability improvement, multi-solution reduction, and nonlinear solving method, and tries to improve the feasible of inverse proble solution based link analysis method in solving actual social network problems by solving corresponding theoretical problems. The main innovations include: propose syetem model construction methods based on nonlinear fitting and supervised machine learning; propose inverse proble solution methods based on damping least square, constrained optimization, and multiple object optimization to improve stability; propose inverse problem solution method based on estimating iteration initial value using stochastic simulation an

英文关键词： social network；link analysis；multiple solutions；stability；nonlinearity