Semi-Automated Processing of Interconnected Dyads Using Entity Resolution (SPIDER)

 描述（由申请人提供）：本提案的总体目标是开发和验证一种新的创新软件系统，以帮助研究人员更严格地构建社交网络。在过去的二十年中，越来越多的研究采用社会网络分析（SNA）来了解艾滋病毒和性传播感染（STI）的传播。绘制“风险网络”图-其中个人通过传播感染的纽带联系在一起-对艾滋病毒和性传播感染的行为流行病学产生了特别有价值的见解，并为针对艾滋病毒高危人群或艾滋病毒感染者设计的有希望的干预措施提供了信息。然而，尽管取得了这些进展，基于国家统计局的艾滋病毒/性传播感染研究也日益流行，但主要的方法和技术挑战阻碍了该领域的进一步进展。SNA催化重大流行病学进展的能力依赖于研究人员的能力， 从行为数据中构建参与者网络的有效表示。建立风险网络或确定参与者及其伙伴之间的直接和间接关系的标准协议涉及将参与者的姓名和人口统计数据与所提供的有关指定伙伴的数据进行匹配。识别和匹配网络中的重复个体的过程（即，“实体解析”[ER]）通常通过费力的手动交叉引用过程来进行。这些程序的可重复性有限，并可能导致网络结构的错误指定。有效网络构建的进一步复杂化是ER标准：（1）未正式化;（2）在不同环境和人群中的研究中有不同的规定;（3）很少在已发表的文献中解释。将联合收割机强大的自动ER流程与定制和定性输入能力相结合的半自动化工具有可能显著提高风险网络构建的速度和准确性。目前用于健康研究的ER工具往往集中在可用ER技术的静态子集上（例如，人口统计学中的相似性，基于语音的匹配技术）而不结合现有技术的方法（例如，机器学习）。拟议的软件，半自动化处理互联二进制使用实体解析（SPIDER），将为用户提供一个系统，使高效，半自动化网络建设使用强大的，统计上严格的ER算法，丰富的桌面注释工具库，和安全的基于Web的技术。天基信息平台的可定制性将允许在使用不同设计的研究中使用多学科工具，并将包括专门针对艾滋病毒/性传播感染研究中新出现的方法趋势的创新功能。该项目的总体目标是提高研究中使用的网络建设的效率和质量，从而改善基于网络的干预措施的证据基础，以减缓艾滋病毒的传播。

 DESCRIPTION (provided by applicant): The overarching aim of this proposal is to develop and verify a new and innovative software system that will assist researchers in more rigorously constructing social networks. During the past two decades, studies have increasingly employed social network analysis (SNA) to understand HIV and sexually transmitted infections (STI) transmission. The mapping of "risk networks," in which individuals are connected by infection-spreading ties, has yielded especially valuable insight into the behavioral epidemiology of HIV and STI, and has informed promising interventions designed for people at risk for or living with HIV. However, despite these advances and the burgeoning popularity of SNA-based HIV/STI research, major methodological and technological challenges are hindering further progress in the field. SNA's ability to catalyze major epidemiologic advances relies on researchers' ability to construct valid representations of participants' networks from behavioral data. The standard protocol for constructing risk networks, or identifying direct and indirect relationships among participants and their partners, involves matching participants' names and demographics with data provided about named partners. This process of identifying and matching duplicate individuals in the network (i.e., "entity resolution" [ER]) is often conducted through laborious, manual cross-referencing procedures. These procedures are limited in their reproducibility and may lead to misspecification of network structure. Further complicating valid network construction is that ER criteria are: (1) not formalized; (2) specified differently across studies n various settings and populations; and (3) rarely, if ever, explained in the published literature. Semi-automated tools that combine powerful automated ER processes with capacities for customization and qualitative input have the potential to dramatically improve the speed and accuracy of risk network construction. Current tools for ER in health research tend to focus on a static subset of available ER techniques (e.g., similarity in demographics, phonetic-based matching techniques) without incorporating state- of-the-art approaches (e.g., machine learning). The proposed software, Semi-automated Processing of Interconnected Dyads using Entity Resolution (SPIDER), will provide users with a system that enables efficient, semi-automated network construction using a library of robust, statistically rigorous ER algorithms, rich desktop-based annotation tools, and secure web-based technologies. The customizability of SPIDER will allow for multi-disciplinary utility in studies using varying designs and will include innovative features that specifically respond to emerging methodological trends in HIV/STI research. The overarching goal of this project is to improve the efficiency and quality of network construction used in research, thereby improving the evidence base for network-based interventions that mitigate the spread of HIV.