基于链接重要性和数据场的链接预测算法

发布时间：2018-05-14 07:47

本文选题：链接权重 + 数据场　；参考：《内蒙古大学》2014年硕士论文

【摘要】：随着信息技术的发展,产生了大规模的网络数据,这为进行大规模的网络分析研究提供了充足的数据。近几年网络挖掘的研究迅速崛起,并发展成为一个很热门的研究领域。链接预测是网络分析的一个重要部分,是一个具有挑战性的研究方向。本文围绕网络数据挖掘领域,针对链接预测任务展开了深入的研究。在对现有链接预测算法分析的基础上,重点研究了基于结构相似性的链接预测算法。针对现有基于结构相似性的链接预测方法忽略了网络拓扑本身链接强度的信息,带权的拓扑路径方法中权值较难确定等缺陷,提出基于链接重要性和数据场的链接预测算法。该方法将所有链接边赋予不同的链接权重,同时考虑潜在链接节点间的相互影响,对部分没有链接的节点进行链接预估计,最后利用数据场势函数计算两节点间的相似值。实验结果表明,该方法整体上提高了预测准确性,且参数确定简单,有很高的实用价值。另外针对网络通常是动态变化,且网络规模通常很大,而在应用中,实时性要求又很高,现有的算法复杂度更新代价又较高,难以达到实时要求的现状,提出了网络的特定存储方式以及增量计算方法,达到低代价更新网络的目的。
[Abstract]:With the development of information technology, large-scale network data is produced, which provides sufficient data for large-scale network analysis. In recent years, the research of network mining has risen rapidly, and has become a very popular research field. Link prediction is an important part of network analysis and a challenging research direction. This paper focuses on the research of link prediction task in the field of network data mining. Based on the analysis of existing link prediction algorithms, a link prediction algorithm based on structural similarity is studied. The existing link prediction methods based on structural similarity ignore the information of the link strength of the network topology itself and the weight value of the weighted topological path method is difficult to determine. A link prediction algorithm based on link importance and data field is proposed. In this method, all link edges are given different link weights, and the interaction between potential link nodes is taken into account, and some nodes without links are pre-estimated. Finally, the similarity between two nodes is calculated by using the potential function of the data field. The experimental results show that the method improves the accuracy of prediction, and the parameter determination is simple, and has high practical value. In addition, the network is usually dynamic, and the network scale is usually very large, but in the application, the real-time requirement is very high, the cost of updating the existing algorithm complexity is high, it is difficult to meet the real-time requirements. The special storage mode and incremental computing method are proposed to update the network at low cost.
【学位授予单位】：内蒙古大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP311.13

【参考文献】