声誉在公共物品博弈中的机制研究和多目标云服务推荐方法

发布时间：2018-04-14 09:27

本文选题：空间网络 + 个体声誉　；参考：《天津理工大学》2017年硕士论文

【摘要】：虽然合作行为广泛存在于自然界和人类的社会系统中,可是根据达尔文的生物进化论,合作并不是一个最优的策略。因此,如何解释合作行为的出现在学术界一直是一个热点课题。其中,演化博弈论为解释合作行为的演化过程提供了一个有力的框架。随着云计算技术的迅速发展和大数据时代的来临,在网络中存在着大量的可用服务。因此,用户所关注的重点不再是服务的可用性,而是转移到了服务的质量,现在用户向推荐系统提出了新的要求。为了解释个体声誉在公共物品博弈中所发挥的作用,论文提出了两个新的模型并进行了大量的实验来说明合作行为在格子网络中的演化过程;针对用户的需求,论文设计了两种新的推荐方法来增加服务列表的多样性。论文主要研究内容和创新性研究成果包括以下几个方面:1)提出一种新的考虑个体声誉的演化博弈模型,深入探究空间公共物品博弈合作行为的演化过程。该模型基于个体效用,策略状态根据费米规则进行异步更新,其中个体效用被定义为博弈收益和声誉值幂函数的乘积。大量的仿真实验证明当声誉被引用到效用中,合作行为得到了明显的提高。2)探索引入声誉推理机制后的公共物品博弈在空间网络中的演化博弈。在博弈演化过程中,个体声誉主要由相依网络中上下两层相对应的个体共同决定;在策略更新阶段,根据邻居在两层网络中声誉的平均值、较大值和较小值,选择所要学习的邻居。大量的仿真结果证明,三种声誉计算规则和传统的公共物品博弈模型比较,都能有效地促进合作率的提升。其中,平均值和较小值方案的提升效果要略优于较大值方案。簇的详细形成过程和声誉分布情况说明了三种规则在根据声誉值选择学习个体时的细微差别。3)开发两个新的推荐算法,通过对服务来源地赋权的方法,在保证推荐列表准确性的条件下,进一步提高列表中所推荐的服务多样性,来降低服务推荐不当的风险。实验结果证明论文所设计的推荐算法不仅能够和原有的算法保持相同的时间复杂度,而且能够获得更优化的推荐结果。
[Abstract]:Although cooperative behavior exists widely in nature and human social systems, according to Darwin's theory of biological evolution, cooperation is not an optimal strategy.Therefore, how to explain the emergence of cooperative behavior in academia has been a hot topic.The evolutionary game theory provides a powerful framework for explaining the evolutionary process of cooperative behavior.With the rapid development of cloud computing technology and big data era, there are a large number of available services in the network.Therefore, the focus of the user is no longer the availability of the service, but the quality of the service. Now the user has put forward new requirements to the recommendation system.In order to explain the role of individual reputation in the game of public goods, two new models are proposed and a lot of experiments are carried out to illustrate the evolution of cooperative behavior in grid network.Two new recommendation methods are designed to increase the diversity of service lists.The main research contents and innovative research results include the following aspects: 1) A new evolutionary game model considering individual reputation is proposed to explore the evolution process of cooperative behavior in space public goods game.The model is based on individual utility, and the strategy state is updated asynchronously according to Fermi rule, where individual utility is defined as the product of the power function of game income and reputation value.A large number of simulation experiments show that when reputation is cited into utility, cooperative behavior is significantly improved. 2) exploring the evolutionary game of public goods game in spatial network after introducing reputation reasoning mechanism.In the process of game evolution, the individual reputation is mainly determined by the individuals corresponding to the upper and lower layers in the dependent network, and in the strategy update stage, according to the average value, larger value and smaller value of the neighbor reputation in the two-layer network,Choose the neighbor you want to learn.A large number of simulation results show that, compared with the traditional game model of public goods, the three kinds of reputation computing rules can effectively improve the cooperation rate.Among them, the average value and the smaller value scheme are better than the larger value scheme.The detailed formation process and reputation distribution of clusters show the nuances of the three rules in selecting individuals according to reputation value. Two new recommendation algorithms are developed.Under the condition of ensuring the accuracy of recommendation list, the diversity of recommended services in the list is further improved to reduce the risk of improper service recommendation.The experimental results show that the proposed recommendation algorithm can not only keep the same time complexity as the original algorithm, but also obtain more optimized recommendation results.
【学位授予单位】：天津理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP393.09;TP391.3

【相似文献】