Automating Analysis of Zoning Codes for the National Zoning Atlas

该项目研究了自然语言处理技术，用于从分区代码文本中自动提取信息。本研究的结果有助于提供更好的分区信息，以及数据收集、处理和绘图工具，以阐明分区这一影响我们经济和社会的最重要但尚未得到充分研究的政府权力之一。首先，该项目汇集了数百个分区代码，人工审查它们，并将具体的监管信息纳入可公开访问的国家分区地图集。然后，研究团队开发了自然语言处理方法，将这些分区代码转换为结构化数据，并通过将自动化过程的输出与人工结果进行比较，确保数据的准确性。汇编的数据和开发的工具可以促进具体的、可操作的见解，并开启关于分区对住房供应、交通系统、环境、经济机会、教育机会和我们的食物供应的影响的二次研究。具体而言，该研究考察了当代自然语言处理技术是否能够成功地破译复杂的地方分区代码。该团队收集有关分区代码的信息，并使用这些信息来训练算法，以提取分区代码和法规的关键元素。输出数据由分区变量组成，例如分区区域名称和每个区域的特定规则，以便进行分析和映射。该团队评估了算法的准确性和精度，并与人类执行的分区代码审查进行了对比。如果成功，该数据工具可以为美国各地的政策制定者、研究人员和倡导者提供机会，减少人工工作量，比较他们所在州或全国的分区代码和流程，并为改进决策、参与式规划和区域合作提供高质量的数据。此外，该方法可以指导未来有兴趣为其他类型的行政法文本（包括公路安全手册、建筑规范、城市规划等）创建自然语言处理模型的研究人员。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

This project investigates natural language processing techniques for automating information extraction from zoning code text. The outcome of this research contributes better zoning information, as well as data collection, processing, and mapping tools, to illuminate zoning, one of the most important, yet understudied, governmental powers impacting our economy and society. First, the project assembles hundreds of zoning codes, manually reviews them, and incorporates specific regulatory information into the publicly-accessible National Zoning Atlas. Then, the research team develops natural language processing methods to transform these zoning codes to structured data and ensure data accuracy by comparing the outputs of the automated process to the manual results. The compiled data and developed tools can facilitate concrete, actionable insights and unlock secondary research about zoning’s impact on housing availability, transportation systems, the environment, economic opportunity, educational opportunity, and our food supply. Specifically, the research examines whether contemporary natural language processing techniques can successfully decipher complex local zoning codes. The team collects information about zoning codes and uses this information to train an algorithm for extracting key elements of zoning codes and regulations. The output data consist of zoning variables, such as zoning district names and specific rules for each district, ready for analysis and mapping. The team assesses the algorithm’s performance for accuracy and precision against human-performed reviews of zoning codes. If successful, the resulting data tool can offer local policymakers, researchers, and advocates across the U.S. the opportunity to compare zoning codes and processes in their state or nationwide with decreased manual effort, and to produce quality data for improved decision making, participatory planning, and regional collaboration. Moreover, the approach can guide future researchers interested in creating natural language processing models for other types of administrative law texts, including highway safety manuals, building codes, city plans, and more.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.