【作者】 王启瑞；

【导师】 巩敦卫；

【作者基本信息】 中国矿业大学 ， 控制科学与工程， 2021， 硕士

【摘要】 Github是一种常用的开源软件社区,也是一种特殊的社交网络,存在复杂的社会关系和多样的用户属性。分析该社区用户的影响力,能够识别社区中的关键用户。基于此,社区管理者可以对这些用户施加正面举措,使社区在关键用户的引导下持续健康发展。目前,关键用户主要通过节点重要性度量方法进行识别,由于Github的特殊性,这些算法存在检测效率低下,识别结果不精准的问题。更为重要的是,由于经典的算法没有考虑节点之间的相互作用,不能充分应用于带权网络中。因此,研究有针对性的理论与方法,对Github进行用户影响力评价是非常必要的。本文基于Github中用户自身属性以及不同用户之间的交互关系,对经典Leader Rank算法进行改进,研究Github用户影响力评价方法。本文的主要研究工作如下:(1)针对现有算法只考虑了节点拓扑属性,不能精准应用于Github用户影响力评价,提出一种基于多属性决策和改进Leader Rank算法的Github用户影响力评价方法(MADM＿LR)。该方法基于用户的社会行为特征,设计多属性决策指标,利用改进的Leader Rank算法,计算用户的影响力排名。将所提方法应用于Github用户关注网络,并与已有方法对比。实验结果表明,MADM＿LR既能够识别影响力更强的用户,还有较强的抗干扰性。(2)针对已有复杂网络节点重要性度量方法,不能充分考虑Github用户间的相互作用,提出一种基于用户相似度和改进带权Leader Rank算法的Github用户影响力评价方法(US＿WLR)。首先给出用户相似度和有向网络H指数相关定义。在保证排序结果精准的基础上,结合以上定义进一步构造了带权的Github网络来反映节点间作用强弱。在构造的网络基础上,有效融合节点的初始影响力和网络边权,提出改进带权的Leader Rank算法。将所提方法应用于改造的Github用户关注网络中,实验结果表明,US＿WLR能够识别出传播能力更强的用户。本文结合用户属性提取、网络构建、模型构造与算法求解等多个方面,给出Github用户影响力的评价方法,既能与Github用户的自身特征紧密结合,还能高效精准地识别出影响力更高传播能力更强的用户。本文得到的用户影响力排序结果不仅能用于评估Github社区健康性,还可以用于预测社区的发展趋势。因此,具有重要的理论意义和应用价值。该论文有图14幅,表4个,参考文献85篇。更多还原

【Abstract】 Github is a common FOSS community and a special social network.It contains complex social relations and various user attributes.By analyzing the influence of users in the community,we can identify key users.And community managers can take positive measures to these users,so that the community can develop continuously and healthily.At present,key users are mainly identified by node importance measurement.Due to the particularity of Git Hub,these algorithms have the problems of low detection efficiency and low accuracy.Most importantly,those algorithms did not consider the interaction between nodes.The existing method cannot sufficiently apply on the weighted network.It is critical to develop an accurate evaluation method about Github users’ influence.Based on the user’s own attributes and the interaction between different users,this paper improves the classic Leader Rank algorithm and proposes a method to identify key users.The results are shown as follows:(1)In view of the fact that the existing algorithms only consider the topological attributes of nodes and can not be accurately applied to Git Hub users’ influence evaluation,this paper proposes a evaluation method based on multi-attribute decision-making and improved Leader Rank algorithm(MADM＿LR).This method based on users’ social behavior,provide multiple attributes targeting to user’s influence and calculate the user’s influence rank with the improved algorithm.The proposed method is applied to Git Hub and compared with the existing methods.Experimental results show that MADM＿LR can not only identify more influential users,but also have better anti-interference performance.(2)In view of the fact that the existing node importance measurement methods in complex networks can’t fully consider the interaction between users,this paper proposes a Git Hub users’ influence evaluation method based on users’ similarity and improved weighted Leader Rank algorithm(US＿WLR).First,we give users’ similarity and H-index for directed networks.The weighted network is constructed to reflect the interactions between nodes based on the accuracy of sorting results.On the basis of the constructed network,this paper proposes an improved weighted Leader Rank algorithm which effectively integrates the initial influence of nodes and the network edges’ weight.The proposed method is applied to the modified users’ follow network,and the results show that US＿WLR can identify users with stronger propagation ability.Combined with user attribute extraction,network construction,model construction and algorithm solving,we give the evaluation method of Git Hub users’ influence.This method can not only combine with the characteristics of users,but also identify the users with higher influence and stronger communication ability efficiently and accurately.The ranking results of users’ influence can be used not only to evaluate the health of community,but also to predict the development trend.Therefore,it has important theoretical significance and application value.The thesis has 14 figures,4 tables,and 85 references.更多还原