Constructing and Validating an Automated Coding System for Electronic News Sources

该项目构建、测试和验证了一个开源的自动化系统，用于对来自电子可用新闻来源的社会运动数据进行编码。虽然没有一个来源是完美的，但学者们一致认为，此类数据，特别是关于社会抗议的数据，最好的总体来源是大量新闻来源的汇编。这个项目利用了计算机科学和统计学在机器学习方面取得的进展，并将它们与我们作为社会学家对新闻来源中集体行动事件的识别和编码问题的深入实质性知识相结合。以《纽约时报》为基础的现有备受推崇的手工编码数据集被用作机器学习算法的参考--被称为“分类器”--以识别新闻文章中动作事件的元素并提取相关信息。我们还收集并手工编码从其他地区、国家和国际新闻来源获得的新数据，以提供额外的培训集，以增加我们能够检测到的抗议活动的范围和种类。该项目的一个补充部分为参与收集和编码这些新数据的本科生提供了研究经验。该项目构建、测试和验证了一个开源的自动化系统，用于对来自电子可用新闻来源的数据进行编码。它推进了社会科学中数据收集的状态，并采用了计算机科学和统计学中自然语言处理和有监督机器学习的最新发展。其结果将是一个开源的、公开可用的系统，可以被其他研究人员使用，并进一步改进和扩展。该项目承诺为社会科学家提供一个重要的新方法工具，具有广泛的跨学科价值，并为更有效地汇编集体行动数据打开大门-来自新闻来源的数据可以改善学术研究和公共政策。该项目的所有代码将在开放源码许可下发布，并可通过公共源代码存储库公开访问。这本书将对社会运动、国际关系和外交政策的学者来说是可读和有用的。这项工作也将对大量非学术界人士有用，例如外交政策分析家和决策者、记者以及那些对文本分析和分类的计算方法感兴趣的人。对来自新闻来源的集体行动数据进行更有效和准确编码的能力具有广泛的政策适用性。

This project builds, tests, and validates an open-source automated system for coding social movement data from electronically available news sources. Although no source is perfect, scholars agree that the best general source of such data, specifically that on social protest, is a compilation of a large number of news sources. This project draws on the advances in machine learning developed in computer science and statistics and combining them with our deep substantive knowledge as sociologists of the problems of identifying and coding collective action events in news sources. An existing highly-regarded hand-coded data set based on the New York Times is used as the reference for "training" machine learning algorithms -- called "classifiers" -- to recognize elements of an action event in a news article and extract relevant information. We also collect and hand-code new data drawn from other regional, national, and international news sources to provide additional training sets to increase the range and variety of protests we are able to detect. A supplement to the project provides research experience for undergraduates who will be involved in collecting and coding these new data.This project builds, tests, and validates an open-source automated system for coding data from electronically available news sources. It advances the state of data collection in social science and employs the latest developments in natural language processing and supervised machine learning within computer science and statistics. The result will be an open-source, publicly available system that may be used by other researchers and further improved and expanded.This project promises to provide an important new methodological tool of broad interdisciplinary value to social scientists and to open the door to more efficiently compiling collective action-data from news sources that can improve both academic scholarship and public policy. All of the code for this project will be released under an open-source license and publicly accessible through a public source code repository. This work will be accessible and useful to scholars in social movements, international relations, and foreign policy. The work will also be of use to a large number of non-academics, such as foreign policy analysts and decision-makers, journalists, and those interested in computational methods of textual analysis and classification. The ability to code collective action data more efficiently and accurately from news sources has broad policy applicability.