CHS: Small: Large-Scale Examination of the Impact of Shocks on Crowd Attributes and Performance in Collaborative Volunteering Systems

这个项目将促进对在线资源志愿者贡献者群体在面对与其工作相关的突发、意外事件时的表现的了解。这些志愿者贡献者(统称为“群众”)创造了有价值的资源，如维基百科文章、公民科学项目数据和开源软件。然而，虽然已经有关于人群和情况的哪些方面导致高质量资源的先前研究，但这种研究通常假设人、群体，特别是情况是相对稳定的。在实践中，情况经常会遇到突然的变化或“震惊”：名人或世界事件的逝世可能会影响维基百科上与之相关的文章，而软件项目可能会发布新版本或发现关键错误。在这个项目中，调查人员将使用存储在GitHub网站上的维基百科文章和开源项目的公共历史来分析人群对冲击的反应，以及这如何影响他们创造的资源。为了做到这一点，他们将使用个人和群体行为理论来衡量人群和他们所处理的资源的有意义的属性，然后使用数据分析技术来了解(1)人群在冲击期间如何变化；(2)人群的哪些属性预测了面对冲击时的高弹性和资源质量；以及(3)这些影响如何根据所经历的冲击的类型而变化。从这项工作中获得的见解还将导致为管理软件和社区的人提供设计建议，这些软件和社区使人群能够创造这些有社会价值的资源。这项工作将从构建人群及其产生的资源的特征开始。重点将集中在维基百科之前的经验工作和组织理论工作表明将与协作众包系统中的性能相关的特性。对于人群来说，这些因素包括关于团队组成和参与行为的元素，包括经验、多样性和工作平衡；关于围绕创造资源的团队协调的数量和基调；以及关于基于个人之间的沟通模式推断的协作者的网络结构。在资源的情况下，它包括属性，包括其内部和外部的受欢迎程度，以及评级或估计的当前质量。调查人员将利用这些特征来分析人群在面临各种冲击时的表现，包括资源质量、工人身份和相关外部事件的变化。要做到这一点，调查人员将开发算法来检测人群经历过震惊的时间，然后使用资源质量和人群规模等属性的倾向性分数匹配，找到没有经历过震惊的相似人群的比较集。然后，他们将使用机器学习分类器和分段回归分析技术来分析冲击发生后资源周围人群的组成和行为的变化，相对于比较资源周围的人群行为。一旦开发了这些模型，该团队将把它们应用于预测人群对冲击的预期弹性和协作系统内部可能发生的冲击的问题。

This project will advance understanding of how groups of volunteer contributors to online resources perform in the face of sudden, unexpected events related to their work. These volunteer contributors (collectively, "crowds") produce valuable resources such as Wikipedia articles, data for citizen science projects, and open source software. However, though there has been prior research on what aspects of crowds and situations lead to high quality resources, such research normally assumes that the people, groups, and especially situations are relatively stable. In practice, situations often encounter sudden changes, or "shocks": the death of a celebrity or a world event can affect Wikipedia articles related to it, while a software project might release a new version or discover a critical bug. In this project, the investigators will use the public history of Wikipedia articles and open source projects stored on the GitHub website to analyze how crowds react to shocks, and how that affects the resources they create. To do this they will use theories of individual and group behavior to measure meaningful attributes of both the crowds and the resources they work on, then use data analysis techniques to understand (1) how crowds change during shocks; (2) what attributes of crowds predict high resilience and resource quality in the face of shocks; and (3) how these effects change depending on the type of shock that is experienced. The insights gained from the work will also lead to design recommendations for people who manage the software and communities that enable crowds to create these socially valuable resources.The work will start by constructing features of crowds and the resources they produce. The focus will be on features that prior empirical work in Wikipedia and work from organization theory suggest will be relevant to performance in collaborative crowdsourcing systems. For crowds, these include elements about team composition and participation behavior including experience, diversity, and work balance; about the amount and tone of team coordination around creating the resources; and about the inferred network structure of the collaborators based on individuals' communication patterns with each other. In the case of resources, it includes attributes including their internal and external popularity, and rated or estimated current quality. The investigators will use these features to analyze how the crowds perform when facing a variety of kinds of shocks, including changes in resource quality, worker status, and relevant external events. To do this, the investigators will develop algorithms to detect times when a crowd has experienced a shock, then use propensity score matching on attributes such as resource quality and crowd size to find comparison sets of similar crowds that have not experienced a shock. They will then use machine learning classifiers and segmented regression analysis techniques to analyze changes in the composition and behavior of the crowd around resources after shocks occur, relative to how the crowd behaves around the comparison resources. Once these models have been developed, the team will apply them to questions of predicting both the anticipated resilience of a crowd to a shock and the potential occurrence of shocks internal to the collaboration system.