Developing a Census Based Generative Geodemographic Classification System

利用当代人工智能(AI)的力量，该项目旨在彻底改变我们建立和使用地理人口分类的方式。这将通过更准确地表示社会空间结构和降低基于普查的分类发展的障碍来做到这一点。它还提出了一个用户友好的在线工具，使任何人都可以轻松地创建自己的量身定制、随时可供研究的基于普查的地理人口数据产品。地理人口分类提供了生活在小地理区域内的人口的复杂和多维特征的有用和与政策相关的表示。自20世纪70年代以来，使用人口普查数据的组成部分创建了分类，2001年、2011年和2021年是值得注意的例子，当时英国国家统计局与学术伙伴共同编制了英国第一个开放的地理人口分类。这些“产出区域分类”(OAC)得到了广泛的应用，并启发了伦敦(LOAC)等特定地理区域的本地化模型。然而，自20世纪70年代以来，用于建立地理人口分类的核心方法一直保持着合理的静止，只有轻微的更新。此外，分类的创建也仍然是一个合理的技术过程，限制了其他人产生自己的分类的能力，无论是针对地方还是特定目的。这一提议认为，人工智能，特别是深度学习和机器学习的最新发展，显示出从根本上改变地理人口分类的力量和效用的巨大潜力。首先，通过创造更准确的社会空间结构表征；第二，通过改进的地理人口信息系统显著减少开发新分类的障碍目标和目标本项目的目标是通过将人工智能集成到以下方面来更新用于构建基于人口普查的地理人口分类的现有方法：更自动化地开发输出区水平的输入测量，更好地解释变量之间的非线性地理关系。支持自动描述聚类的工具。支持创建新的面向公众的在线地理人口分类系统，从而能够创建基于人口普查的自定义分类。这将通过以下目标实现：评估自动编码器作为减少输出数据的新方法的使用地区一级的地理人口输入措施。开发一个可操作的机器学习管道，将输出地区一级的人口普查输入通过集群创建。利用大型语言模型(LLM：如集成到ChatGPT中)，开发一个自动化的地理人口统计描述工具，能够对集群特征进行准确的文字描述。制作一个面向公众的在线工具和附带的培训，指导用户创建自己的可供研究的、以普查为基础的地理人口统计数据产品。

Leveraging the power of contemporary Artificial Intelligence (AI), this project aims to revolutionize the way in which we can build and use geodemographic classifications. This will do so by enabling more accurate representations of socio-spatial structure and lowering barriers to census based classification development. It also proposes a user-friendly online tool that will allow anyone to easily create their own tailored, research-ready census-based geodemographic data product.Geodemographic classifications provide useful and policy-relevant representations of the complex and multidimensional characteristics of populations living within small geographic areas. Classifications have been created using components of census data since the 1970s, with notable examples in 2001, 2011 and 2021 when the ONS co-produced the first open geodemographic classifications for the UK with academic partners. These "Output Area Classifications" (OAC) have garnered wide use and inspired localised models for specific geographic areas such as London (LOAC).The core methods used to build geodemographic classification have however remained reasonably static since the 1970s, with only modest update. Furthermore, the creation of classifications also remains a reasonably technical process, limiting the ability for others to produce their own classifications, either for localities or specific purposes.This proposal argues that recent developments in AI, and specifically deep learning and machine learning, show great potential to radically transform the power and utility of geodemographic classification. Firstly, through the creation of more accurate representations of socio-spatial structure; and, secondly, through improved geodemographic information systems that significantly reduce barriers to developing new classificationsAims and ObjectivesThe aim of this project is to update the established methods used to build Census based geodemographic classifications through the integration of AI into:The more automated development of output area level input measures that better account for non-linear geographic relationships between variables.A tool to that enables the automated description of clusters.Enabling the creation of a new public facing and online geodemographic classification system that will enable custom census-based classifications to be created.This will be achieved through the following objectives:Evaluating the use of autoencoders as a new method of data reduction for output area level geodemographic input measures.Developing an operational machine learning pipeline that takes output area level census inputs through to cluster creation.Utilising a large language model (LLM: such as integrated into ChatGPT), to develop an automated geodemographic descriptive tool capable of producing accurate textual descriptions of cluster characteristics.Producing a public facing online tool and accompanying training that will guide users to create their own research-ready census-based geodemographic data products.