Horizon Scanning Through Automated Information Prioritisation

众所周知，警察部队正在寻求改进他们的分析和预测工具，以应对新出现的犯罪活动。但是，警察部队记录的和通过公开来源获得的数据的数量、种类和速度意味着，分析这些数据可能是具有挑战性的。目前的努力通常诉诸于人工数据检查，阻碍了对规模和时间问题的理解。尽管存在这些障碍，水平扫描(HS)将成为未来警务的一个重要方面，使警察部队能够识别新出现的威胁，预测迫在眉睫的犯罪，并最终扑灭未来的犯罪方法。从本质上讲，HS活动的目的是帮助警方领先犯罪分子一步。这项研究的目的是开发一种新的自动化技术，可以识别未来犯罪问题的早期预警信号。这项计划分为三个主题：(I)利用警方所知，(Ii)社会人士所知，以及(Iii)传媒所知。每个部分都将建立在前面技术的基础上，以创建一个单一的、全面的HS系统。为了开发与现实世界相关的研究成果并解决目前HS中的封锁，这项研究将与执法机构建立密切的工作关系。为了实现这些目标，研究将从定义研究的宪法主题开始，如新犯罪类型和/或实施方法的特征。这将为以后关于自动检测和监测(新的)犯罪趋势的工作提供信息。随后将审议数据来源，其中将包括为将创建的开放源码情报储存库开发一个框架，以自动收集和存储来自不同来源的大量数据。该储存库将包括各种来源，如新闻报道、社交媒体数据和社区论坛帖子。还正在作出数据共享安排，以获取警务数据。总的来说，收集的数据将反映这项计划的三个主要主题。为了开发该方法，初步将使用简单的合成数据来评估各种自然语言处理技术在提取罪犯在犯罪时表现出的行为和行为的描述方面的有效性。然后将在犯罪报告中复制这一点，以确定在这方面的表现，并促进就警方数据收集做法提出建议。然后，将采用异常检测等机器学习方法(类似于银行用来识别欺诈交易的方法)，以测试提取的信息在监测犯罪变化方面的潜力。这项研究的下一部分将利用OSINTR和机器学习(计算机从数据中学习模式，而无需明确编程)，以及不同的数据类型，以进一步推广早先在主题提取和检测中使用的方法。这里的分析将谈到社会和媒体对正在出现的和未来的犯罪问题所了解的主题。然后，这项研究将通过利用不同的数据源和使用自我学习算法来结束对预测能力的最终研究(例如，预测明天的犯罪标题)。这样做的目的是生成简短的问题片段，以便分析师进一步研究。研究的主要目标是开发一个自动化的HS系统，该系统可以自动整理各种开放数据源的信息可以从不同的数据类型中提取与犯罪相关的信息监视犯罪和实施方法的异常变化可能会以早期警告的形式对未来的问题做出预测与此研究相关的EPSRC研究领域是人工智能技术信息系统自然语言处理操作研究

It is no secret that police forces are seeking to advance their repertoire of analytical and predictive tools to deal with emerging crimes.However, the volume, variety, and velocity of data recorded by police forces and available through open sources means that analysing such data can be challenging. Current efforts typically resort to manual data inspection impeding understanding of problems at scale and over time. Despite these obstacles, horizon scanning (HS) is set to become a prominent aspect of future policing, by enabling police forces to identify emerging threats, anticipate imminent crimes and to ultimately extinguish future methods of perpetration. In essence, HS activities aim to aid the police in staying one step ahead of criminals. The objective of this research is to develop a novel automated technology that can identify early warning signals of future crime problems. The project is split into three main themes: (i) utilising what the police know, (ii) what the community knows, and (iii) what the media knows. Each part will build on techniques from the previous, to create a single, comprehensive, HS system. In order to develop research output that is relevant in the real world and to resolve current blockades in HS, this research will foster close working relationships with law enforcement agencies.To achieve the objectives, the research will begin with a conceptual piece that defines the constitutional motifs of the research such as the characterisation of new crime types and/or methods of perpetration. This will inform subsequent work on automatically detecting and monitoring (novel) crime trends. Data sources will then be considered and will include developing a framework for an open-source intelligence repository (OSINTR) to be created to automatically collect, and store, a large quantity of data from different sources. The repository will include diverse sources such as news reports, social media data and community forum posts. Data sharing arrangements are also being made to gain access to policing data. Overall, the collected data will reflect the three main themes of this project.To develop the method, simple synthetic data will initially be used to evaluate various natural language processing techniques for their usefulness in extracting descriptions of the actions and behaviours exhibited by an offender whilst committing a crime. This will then be replicated on crime reports to establish performance in this context and facilitate recommendations for police data collection practices. Machine learning methods such as anomaly detection (similar to those used by banks to identify fraudulent transactions) will then be implemented to test the extracted information's potential for monitoring changes in crime. The next part of the research will utilise the OSINTR and machine learning (where computers learn about patterns from the data, without being explicitly programmed), and different datatypes to further the methods used earlier in theme extraction and detection. The analysis here will speak to the themes of what the community and the media know about emerging and future crime problems. The research will then conclude with a final study on predictive ability (e.g., anticipating tomorrow's crime headlines today) by utilising the different data sources and employing self-learning algorithms. The purpose of this is to generate short problem snippets that analysts can then investigate further.The primary objectives of the research are to develop an automated HS system thatAutomatically collates information from a variety of open data sourcesCan extract crime related information from different data typesMonitors anomalous changes in crime and methods of perpetration Can potentially make predictions in the form of early warnings concerning future problems The EPSRC research areas relevant to this research are AI technologiesInformation systemsNatural language processingOperational Research