Implementing and Evaluating a Machine Learning Tool for Entity Resolution in Drug Use and Sexual Contact Networks of YMSM

项目摘要 将接触网络数据与分子监测数据相结合的努力为艾滋病毒提供了巨大的希望 跟踪和干预。然而，缺乏促进综合分子-社会监测的工具 仍然是取得进展的重大障碍。例如，大多数联系人网络数据仅包含有关以下内容的信息 一个人的直接性伴侣和吸毒者。然而，相同的合作伙伴可能会出现在 多个人的联系网络。因此，合作伙伴必须在联系网络中进行匹配-a 称为实体解析(ER)的流程-以便提供整个联系网络的准确视图 结构。急诊室流程目前需要大量资源来手动匹配个人，或者 在编程方面拥有丰富的技术专长，以便使用概率法更有效地匹配个人 模特们。因此，这个项目将1)开发一种机器学习算法来匹配不同的个体 个人联系网络，并使用现有的大量与男性发生性关系的年轻男子的数据集进行验证， 以及2)创建图形用户界面以将算法实现为现有工具的附加包 用于网络数据捕获和处理(Network Canvas)。该项目的成果将提供一个开放的-- 来源和免费提供的工具，可以极大地减少跨联系人匹配个人的障碍 网络，从而为研究人员和公共卫生官员提供不受阻碍的访问 毒品使用和性网络的基本结构，以及整合联系网络数据的有力工具 进行分子监测。

Project Summary Efforts to integrate contact network data with molecular surveillance data provide enormous promise for HIV tracking and intervention. However, the lack of tools to facilitate integrated molecular-social surveillance remains a substantial barrier to progress. For example, most contact network data only contains information on the immediate sexual and drug use partners of a single individual. Yet, the same partners can appear across the contact networks of multiple individuals. Therefore, partners must be matched across contact networks - a process called entity resolution (ER) - in order to provide an accurate view of the overall contact network structure. The process of ER currently requires either substantial resources to manually match individuals or considerable technological expertise in programming to more efficiently match individuals using probabilistic models. Accordingly, this project will 1) develop a machine learning algorithm to match individuals across personal contact networks and validate it using a large existing dataset of young men who have sex with men, and 2) create a graphical user interface to implement the algorithm as an add-on package to an existing tool for network data capture and processing (Network Canvas). The results of this project will provide an open- source and freely available tool that can drastically reduce barriers to matching individuals across contact networks, thereby providing researchers and public health officials with unencumbered access to the underlying structure of drug use and sexual networks, and a potent tool for integrating contact network data with molecular surveillance.