社会经济系统的空间结构与动力学研究
Research on the Spatial Structure and Dynamics of Socio-Economic Systems

社会经济系统是一类重要的复杂系统,涉及到人类经济活动与所处社会环境的复杂相互作用。人类的认识和行为不断发生变化,主观决策过程极大地影响社会经济系统的运行。精准和及时地感知社会经济态势,揭示和理解社会经济发展规律,有重大的理论意义和应用价值。洞察社会经济发展中各方面的状态,并对其发展趋势进行准确的预测,有助于科学地指引社会经济决策。揭示个体的社会经济行为模式,能帮助逐渐实现预测性管理。刻画宏观的社会经济结构,有助于探寻经济发展路径。如何有效地分析社会经济系统的结构与演化规律,是多学科交叉研究领域所关注的重要科学问题,近年来得到了包括计算机科学、网络科学、复杂性科学、统计物理和社会经济学在内的很多相关学科的极大关注。传统的社会经济研究依靠定性或半定量方法,导致不容易从机制层面认识相关问题。利用传统普查数据计算宏观经济指标,整个过程不但消耗大量资源,而且时间滞后很长。不仅如此,传统分析方法难以洞察经济发展的结构转变,无法刻画经济发展过程中的复杂性,缺乏预测经济发展趋势的能力。近年来,硬件和技术的同步发展推动数据化浪潮,为社会经济研究带来了前所未有的机遇和改变。数据获取方式的进步,提高了大规模社会经济数据的可用性。数据规模和多样性的增加,促进了社会经济分析工具和方法论的变革。逐渐应用的新数据和新方法,提高了社会经济研究的定量化程度,催生了一个新兴的交叉学科研究分支,称为计算社会经济学。本文在计算社会经济学框架下,将分别从微观、中观和宏观层面研究社会经济系统的状态推断和结构建模,进而以理论结合实证的方式探究经济的结构演化和发展策略。特别地,不同层面的研究基于类似的空间网络结构和动力学理论基础。本文主要的研究内容和创新点总结如下:(1)在微观层面,基于非干预行为数据研究了社会经济预测性管理。通过分析匿名校园卡数据,提出了谨严性指数来刻画个体行为规律程度。发现谨严性与学生成绩显著相关,使用谨严性特征能显著提高排序学习算法对学生成绩的预测效果。基于企业社会化平台数据构建互动网络和社会网络,发现利用员工在网络中的位置能预测其升离职的可能性。特别地,互动网络比社会网络的预测能力强,预测离职比预测升职容易。通过分析大规模在线平台数据,以量化方式揭示了一些社会经济现象,包括团队规模在8人以下能提高员工沟通效率和绩效表现,中国社交圈规模也在邓巴数150人左右,职场中存在身高溢价和性别不平等现象。(2)在中观层面,基于在线用户评分数据研究了社会经济系统排序。针对信誉排序问题,提出了基于群组的信誉排序GR算法。不依赖传统的产品质量假设,GR算法根据评分群组规模计算用户信誉。真实数据集上的实验结果表明,GR算法对用户的信誉排序比基准算法更准确。进一步,利用迭代寻优过程改进GR算法,提出了迭代信誉排序IGR算法。同时考虑用户数量和信誉计算群组规模,IGR算法的排序准确性和鲁棒性更好。针对产品排序问题,提出了节点相似性CosRA指标,基于此提出的CosRA推荐算法表现更好。进一步,提出了考虑用户信任关系的CosRA+T推荐算法,发现过度依赖信任关系有损推荐效果。(3)在宏观层面,基于大规模真实数据刻画和分析了社会经济结构。利用企业注册信息数据,刻画了中国区域经济复杂性。发现复杂性ECI指标和Fitness指标对中国区域经济发展的预测能力相当,复杂性与收入不平等性负相关。利用人力和企业数据,分别构建了巴西和中国区域产业空间。发现两者都有“核心-边缘”结构,复杂程度高和低的产业分别占据核心和边缘位置。中国产业空间还有“哑铃型”结构,在时间演化上存在区域竞争。利用微博和简历数据,分别构建了信息和人才流动网络。发现根据网络的结构特征能推断区域经济发展水平。特别地,人才流动网络的预测能力强,结合两个网络的特征能解释大约84%的GDP差异。(4)在经济发展和结构演化方面,基于空间网络研究了经济演化路径和产业升级策略。利用空间网络模型和传播动力学过程,揭示了网络的空间结构对信息传播的影响。发现空间网络的长边分布能改变靴襻渗流的相变类型,长边分布的幂指数-1为出现相变点不变的双相变的临界值。针对产业空间和地理近邻网络,分别提出了经济发展的相似技术学习途径和近邻区域学习途径。发现两条途径都能促进区域发展新产业,但两者存在替代效应。进一步,以理论分析结合实证数据,研究了发展经济的最优策略。发现缩短距离能提高协同学习效果,引入高铁能提高区域的产业相似性和生产率,两条协同学习途径都存在最优发展策略。计算社会经济学是一个新兴研究分支,在数据和方法上面临新挑战和新机遇。在未来研究中,值得进一步探索社会经济系统的空间结构与动力学,提高对社会经济态势的感知和对发展规律的理解。长期而言,数据驱动的研究范式必将成为解决社会经济问题的主流方法论,也将深刻地改变社会经济研究的图景。

Socio-economic systems are an important branch of complex systems,which involves the complex interactions between people’s economic activities and the social environment in which they live.With the constant change of cognition and behavior,people’s subjective decision-making process greatly affects the operation of socio-economic systems.To accurately and timely perceive socioeconomic situation and to reveal and understand the law of socioeconomic development have great theoretical and practical values.Revealing the status of socioeconomic development in many aspects and predicting the development trends with desirable accuracy can greatly help to guide socioeconomic decision-making.Uncovering the socioeconomic behavioral patterns of individuals can contribute to gradually realizing predictive management.Quantifying the macro socioeconomic structure can help to explore the path of economic development.How to effectively analyze the structure and evolution of socio-economic systems is an important scientific issue in the interdisciplinary research field,and it has recently received great attention from many related disciplines including computer science,network science,complexity science,statistical physics and socioeconomics.Traditional socioeconomic research relies mainly on qualitative or semi-quantitative methods,which makes it difficult to understand relevant issues at the mechanism level.The process that calculates macroeconomic indicators based on traditional census data not only consumes substantial resources,but also follows a long-time delay.Besides,traditional analytical methods have difficulty in tracking the structural transformation of economic development,fail to quantify the complexity of economic development and are lack of predictive power on development trends.The recent simultaneous development of hardware and technology is driving a new wave of big data,which has brought unprecedented opportunities and changes to socioeconomic research.The advances in methods of data acquisition have increased the availability of large-scale socioeconomic data,and the increases in the size and diversity of data have contributed to the transformation of socio-economic analytical tools and methodologies.The application of novel data and methods has gradually increased the level of quantification in socioeconomic research and led to the emergence of a new scientific branch,named Computational Socioeconomics.Under the framework of computational socioeconomics,this dissertation will investigate the status inference and structural modeling of socio-economic systems from the micro,meso and macro levels,and explore the evolution of economic structure and the optimal strategy for economic development through theoretical and empirical studies.In particular,studies at different levels are based on the similar theoretical basis of network spatial structure and dynamics.The main contents and major contributions of this dissertation are summarized as follows:(1)At the micro level,the predictive management of socio-economic systems was studied based on unobtrusive behavioral data.By analyzing data recorded by anonymized campus cards,we proposed a novel orderliness measure to quantify the regularity of individual behavior.Orderliness is significantly correlated with student academic performance,and it can largely improve the performance of learning-to-ranking algorithm on predicting student academic performance.Based on the analysis of two employee networks built on data from an enterprise socialization platform,we found that the locations of employees in both networks are predictive to the possibility of their promotion and resignation.In particular,action network has stronger predictive power than social network,and predicting resignation is easier than predicting promotion.Moreover,by analyzing large-scale online platform data,we revealed some socio-economic phenomena in a quantitative way,including keeping team size below 8 can improve employee’s communication and performance,the size of Chinese social circle is also around Dunbar’s Number 150,and there are height premium and gender inequality in the workplace.(2)At the meso level,the ranking of socio-economic systems was studied based on online user rating data.To solve the of problem reputation ranking,we proposed a group-based reputation ranking(GR)method.Instead of relying on the traditional assumption of product quality,GR method calculates user reputation based on the size of rating groups.Experiments based on real-world datasets showed that GR method outperforms benchmark methods in the accuracy of ranking users by their reputation.By introducing an iterative process into the GR method,we further proposed an iterative group-based ranking(IGR)method.Considering both the number and the reputation of users when calculating the group size,GR method exhibits better accuracy and robustness in reputation ranking.To solve the problem of object ranking,we proposed a novel vertex similarity measure,named CosRA index,based on which we developed a CosRA-based recommendation algorithm that exhibits better performance.Further,we proposed a trustbased recommendation algorithm,named CosRA+T,and found that relying too much on trust relations among users is detrimental to recommendation performance.(3)At the macro level,socio-economic structures were quantified and analyzed based on large-scale real data.Using firm registration information data,we quantified China’s regional economic complexity.We found that ECI index and Fitness index exhibit comparable predictive power for China’s regional economic development,and economic complexity is negative correlated with income inequality.Using labor and firm data,we built Brazil’s and China’s regional industry space,respectively.We found that both industry spaces exhibit a “core-periphery” structure,where industries with high and low level of sophistication occupy the core and the periphery of the industry space,respectively.Moreover,China’s regional industry space has a “dumbbell” structure,and its time evolution has regional competitions.Based on Weibo and resume data,we built information flow and talent mobility network,respectively.We found that regional economic status can be inferred from the structure of both networks.In particular,talent mobility network exhibits a stronger predictive power,and combining the structures of both networks can explain about 84% of the variance in GDP.(4)In economic development and structure evolution,the path of economic evolution and the strategy of industrial upgrading were studied based on spatial networks.By leveraging the spatial network model and the spreading process,we revealed the effects of the spatial structure of networks on information diffusion.We found that the distribution of long-range links of spatial networks can change the phase transition of bootstrap percolation,where the exponent-1 of the distribution of long-range links is a critical value for the presence of a double phase transition with two nearly constant critical points.For industry space and geographical adjacent networks,we proposed the inter-industry learning and the inter-regional learning for economic development,respectively.We found that both collective learning channels can increase the probability of development new industries,while they exhibit an alternative effect.Moreover,we explored the optimal strategy for economic development using both theoretical and empirical analyses.We found that reducing geographical distance can enhance the collective learning effects,introducing high-speed rail can increase regional industrial similarity and productivity,and both collective learning channels have optimal strategies for industrial development.Computational socioeconomics is an emerging research branch,and it faces new challenges and opportunities in both data and methods.In future studies,it is worthwhile to further explore the spatial structure and dynamics of socio-economic systems,and to improve the perception of socioeconomic situation and the understanding of the law of development.In the long run,data-driven research paradigm will become the mainstream methodology for solving social and economic problems and will profoundly change the landscape of socioeconomic research.

复杂网络; 社会经济系统; 排序算法; 经济复杂性; 网络结构;

complex networks; socio-economic systems; ranking method; economic complexity; network structure;

周涛;

F124;F224

9920512214K