Analytics, AI/ML
January 30, 2024

Improving Policing in US with NLP

Cogent Infotech
Location icon
January 30, 2024

Law enforcement agencies face the challenge of identifying and responding to criminal activity in a timely and effective manner. With advancements in technology, natural language processing (NLP) has emerged as a powerful tool for law enforcement agencies seeking to improve policing and keep communities safe.

NLP is a branch of artificial intelligence (AI) that enables computers to understand, analyze, and generate human language. When combined with problem-oriented policing (POP), NLP can provide valuable insights into criminal activity patterns and help develop targeted interventions to address the root causes of crime.

By automating the analysis of large datasets, NLP can streamline the POP process and enable crime analysts to focus on developing effective solutions to address criminal activity. With organized retail crime causing a staggering $100 billion in annual losses in the United States in 2021, the potential benefits of NLP for law enforcement agencies are enormous. However, it is important to note that NLP is not a silver bullet solution and must be used ethically and transparently.

This blog aims to provide readers with a fresh perspective on how technology can be leveraged to improve policing and build safer communities. By exploring the intersection between NLP and policing, readers will gain insights into the potential benefits of NLP, as well as the ethical considerations that must be taken into account when implementing NLP in law enforcement.

Problem-Oriented Policing (POP)

POP is an innovative approach that seeks to address the root causes of crime and disorder. Traditional policing tends to focus on reactive responses to criminal activity, whereas POP is proactive and solution-focused. By identifying the underlying issues that contribute to crime, law enforcement agencies can develop targeted interventions that address the root causes and sustainably reduce crime.

Following the implementation of the program, an evaluation showed a statistically significant reduction of 63% in the average number of youth homicide victims in the city of Boston. This indicates the effectiveness of the program in addressing the underlying causes of crime and reducing criminal activity in the targeted areas.

By using NLP, law enforcement agencies can take the implementation of POP to the next level. They can use NLP to analyze large volumes of data and identify patterns and insights that inform tailored solutions to complex issues.

As a result, the use of NLP can enhance the effectiveness and efficiency of POP, providing law enforcement agencies with valuable insights to improve their strategies. Policymakers, law enforcement professionals, and community leaders can all benefit from the use of POP and NLP. It's a powerful combination that could help tackle crime and disorder in a more targeted and sustainable way.

The Rationale Behind the POP

In the world of policing, reacting to crime after it happens is no longer enough. That's where Problem-Oriented Policing comes in. Herman Goldstein introduced POP as an alternative approach to traditional policing methods. 

POP seeks to understand the underlying issues that lead to crime and disorder to prevent them from happening in the first place. According to the crime triangle theory, for a crime to occur, a motivated offender, a suitable target, and a setting without a capable guardian must converge. POP aims to disrupt this convergence and prevent crime opportunities from arising. By addressing the root causes of crime, POP provides a proactive and sustainable approach to policing.

The College of Policing conducted research that suggests POP is effective in reducing criminal behavior. The research found that this type of policing contributed to a general decrease in offending, and one study reported a substantial 33.8%  drop in crime and disorder.

The Objectives of POP in Contemporary Policing

POP has two main objectives in contemporary policing:

Preventing crime recurrence: 

POP analyzes patterns behind similar incidents to intervene in the generation process and avoid future occurrences. This proactive approach involves identifying the factors that contribute to crime and addressing them at the source. By understanding the root causes of crime, law enforcement agencies can work collaboratively with communities to implement targeted and effective solutions that prevent crime from happening again.

Reducing opportunities for the crime: 

POP aims to reduce the attractiveness of crime in areas where it is traditionally higher by disrupting the crime triangle. By doing so, this approach seeks to decrease the chances of crime, which can reduce reliance on the criminal justice system and prevent the criminalization of individuals.

The SARA Framework for POP

John Eck and William Spelman developed and coined the SARA Model, which builds on the principles of POP. The SARA framework is a cornerstone of POP, offering a systematic approach to tackling crime and disorder by addressing its root causes. 


  • Identify recurring problems that concern both the public and the police.
  • Understand the consequences of these problems for the community and law enforcement.
  • Prioritize problems based on their impact.
  • Develop broad goals to address identified issues.
  • Confirm the existence of problems through data analysis.
  • Determine the frequency and duration of problems.
  • Select problems for an in-depth examination.


  • Investigate the events and conditions that precede and accompany problems.
  • Identify relevant data to be collected for analysis.
  • Research existing knowledge about the problem type.
  • Evaluate the current response to the problem and identify its strengths and limitations.
  • Narrow the scope of the problem for targeted intervention.
  • Identify resources to aid in understanding the problem.
  • Develop a working hypothesis to explain the problem's occurrence.


  • Brainstorm new intervention strategies.
  • Research successful interventions in other communities with similar problems.
  • Choose the most appropriate interventions.
  • Outline a response plan and assign responsibilities.
  • State-specific objectives for the response plan.
  • Execute the planned activities.


  • Determine if the response plan was implemented effectively.
  • Collect pre- and post-response qualitative and quantitative data.
  • Evaluate whether broad goals and specific objectives were achieved.
  • Identify any new strategies needed to improve the original plan.
  • Conduct an ongoing assessment to ensure sustained effectiveness.

Real-Life POP Example

Washtenaw County Sheriff's Office's Macarthur Blvd project,

One real-life example of the successful implementation of Problem-Oriented Policing (POP) is the Kansas City Police Department (KCPD)'s Risk-Based Policing project, which aims to reduce and prevent violent crime and improve the citizens' quality of life.

Scanning: Kansas City, Missouri, has struggled with high levels of violent crime for a long time. The city's homicide rate has ranked in the top ten for cities with over 250,000 residents for eight out of ten years prior to 2017. From 2015 to 2018, there was an increase in murders, aggravated assaults, armed robberies, subjects actively resisting arrest, and firearms recoveries, making the city more violent for both citizens and police officers.

Analysis: Traditional strategies like hot spot policing or focused deterrence didn't work very well, so the police department looked for a new strategy that would work better. They wanted an approach that would address the root causes of crime through prevention and community engagement, as well as traditional law enforcement methods. To help with this, they decided to use a technique called Risk-Based Policing (RBP), which is based on a type of analysis called Risk Terrain Modeling (RTM).

Response: The RBP initiative began in March 2019 and involved the police department working with community members, municipal departments, and the Rutgers Center on Public Security (RCPS). They targeted thirteen areas for treatment and compared the results to four similar areas that didn't receive treatment. The outcome evaluation showed that target violent crimes decreased significantly by over 22% in the treated areas compared to the non-treated areas after a year (2020). They also found that there were fewer officer-initiated activities resulting in arrests or citations, which suggests that the initiative relied less on strict law enforcement and more on other methods.

Assessment: The RBP initiative reduced violent crime and improved the relationship between the KCPD and the community. By using data as a common language and inviting feedback from citizens, the police department was able to engage in true dialogue with the community. As a result, residents felt heard, and a stronger relationship was forged between the police and the community, including the post-George Floyd era.

Challenges to POP Implementation

POP aims to identify and solve the underlying causes of crime through strategic and informed decision-making. However, the implementation of POP has encountered several challenges, such as:

Organizational Barriers:

One of the major organizational barriers to POP implementation is the lack of analytical skills and heavy reliance on enforcement tactics and other situational crime prevention responses within police agencies. It is criticized as "shallow" problem-solving. The traditional response policing model does not require the same level of analytical expertise, resulting in a shortage of personnel with the necessary knowledge and skills to deliver POP effectively. 

Resource Limitations:

The manual analysis of textual data from various sources, such as crime notes, witness statements, and forensic reports, is time-consuming and labor-intensive. Resource limitations limit the effectiveness of POP, as selective analysis may result in important factors going unnoticed.

Resistance to change:

The shift from traditional response policing to a problem-oriented approach requires a significant change in mindset and organizational culture. Police agencies must address this resistance by engaging with their personnel and effectively communicating the benefits of POP. 

The challenges of implementing POP were highlighted during the COVID-19 pandemic and post-George Floyd killing, where its effectiveness was questioned. While POP has its challenges, its benefits in reducing crime and improving community relations cannot be ignored. Therefore, police agencies must address these challenges and work towards implementing POP successfully.

The Potential of NLP in Addressing Challenges

The application of NLP can address some of the challenges in implementing POP by automating certain processes and reducing the analytical burden on police agencies. Here are a few ways NLP can be integrated with POP:

Automated analysis of police reports:

NLP can automatically analyze police reports and extract relevant information, such as patterns, trends, and relationships between incidents. This can help identify underlying issues and direct resources more efficiently. Additionally, NLP can be applied to automate the analysis of emergency calls to determine the severity of reported incidents. Hence law enforcement agencies can improve their response times and prioritize resources for situations that require immediate attention.

Sentiment analysis of community feedback: 

By analyzing the sentiments expressed by community members through social media, forums, or other platforms, NLP can help police understand public concerns and identify areas where interventions may be most needed.

Improved communication and collaboration: 

NLP can enhance communication between analysts, field personnel, and other stakeholders by automatically summarizing lengthy reports, translating languages, and identifying key information.

Predictive analytics: 

Using NLP and machine learning techniques, law enforcement agencies can develop predictive models to identify potential crime hotspots and allocate resources accordingly. Research confirms that using NLP and Deep Learning techniques in analyzing Police and Health records can help predict future offenses in families and cases of domestic violence.


With the help of NLP-powered interactive simulations and scenarios, officers can be trained to identify and respond to a wide range of complex situations they may encounter in their line of duty. These simulations can be designed to replicate real-world scenarios and provide officers with a safe and controlled environment to practice their problem-solving skills.

The Value of Police Free-Text Data

Police departments often collect free-text narratives called Modus Operandi (MO), which can be incredibly valuable for implementing POP. This data can include information on criminal activity, suspects, and even victim reports. However, utilizing this data can be challenging due to its sensitivity, specialized vocabulary, and resource limitations. Fortunately, NLP can help police departments overcome these challenges and continue improving their policing efforts.

Understanding Police Free-Text Data

Police departments often collect free-text data, which can be a goldmine of information for improving their operations and procedures. But, as with any unstructured text, the unedited nature of police free-text data can pose challenges for NLP models. These models are typically trained on more structured datasets, which means they may struggle to accurately analyze free-text data that contains misspellings, contractions, and varying capitalization rules. Despite these challenges, preliminary experimental work has shown that existing NLP models can still provide sufficient coverage of the language without any adaptation.

Data Sensitivity Challenges

Police free-text data can contain a wealth of valuable information for improving policing efforts. However, this data can also be sensitive, containing personal information that's subject to local laws and regulatory frameworks. As a result, police agencies may face difficulties in sharing this data for academic research or utilizing external resources for analysis.

This is where NLP comes in. NLP provides a solution for police departments to analyze their free-text data while maintaining data privacy and security. For example, in a recent study, NLP methods were applied to free-text scam reports to support crime script analysis and identify points of intervention where scams might be disrupted and prevented.

However, NLP implementations must have low hardware requirements and be easily accessible to practitioners who may not be NLP or machine learning experts.

So, to overcome these challenges, a low-risk approach can be adopted, characterized by:

  • Requesting low-risk data, such as modus operandi information, is meant for sharing within the criminal justice system.
  • Performing in-house pre-processing to redact sensitive information using a white-listing approach.
  • Storing data in secure, compliant environments accessible only by project members.
  • Vetting research team members to ensure they meet necessary standards for data handling.
  • Sharing insights with police partners and obtaining approval from multiple parties before publication.

Related Work

This section explores the related work and research that delves into the use of NLP for enhancing law enforcement efforts.

Crime count analysis:

The police provide numeric crime counts based on reported crimes online, which includes a time stamp, region, and crime category. According to the research, textual time-stamped crime news articles can be hypothesized to provide additional information and context that can be used to explain and predict numeric crime counts and may be useful in explaining inferred patterns of criminal activity based on the counts.

Sentiment analysis for identifying public safety concerns:

research aimed to identify public safety concerns and areas requiring more attention from law enforcement. By monitoring the sentiments expressed on social media platforms, authorities could gain insights into potential criminal activities or community unrest, allowing them to respond proactively and allocate resources more effectively.

Automated analysis of emergency calls:

Another promising research of NLP in law enforcement is the automated analysis of emergency call transcripts. By developing models that can quickly analyze call transcripts and determine the severity of reported incidents, law enforcement agencies can improve their response times and prioritize resources for situations that require immediate attention.

Computer vision approach to crime prediction: 

Computer vision involves teaching computer systems to understand, interpret, and analyze digital images and video footage. A study explores how a combination of ML and computer vision algorithms can help law enforcement agencies detect, prevent, and solve crimes more accurately and quickly. 

NLP Techniques for Enhancing Law Enforcement

The Defense Advanced Research Projects Agency's (DARPA) Deep Exploration and Filtering of Text (DEFT) program uses NLP to extract operationally relevant information from unstructured text data automatically. This technology helps defense analysts derive actionable insights from data and make more informed decisions.

Similarly, in this context, NLP offers numerous applications that can directly benefit law enforcement agencies, particularly in the POP. By reducing the analytical burden associated with real-world police settings, NLP can help streamline processes and enhance decision-making.


Classification is a crucial NLP application that can assist police agencies in understanding contextual factors related to specific offenses. In fact, the advantages of automating security classification through NLP are highlighted by researchers who utilized this technique to classify US security documents. 

Police officers often flag crimes with keywords, such as noting if an offender is under the influence of alcohol or illegal drugs. However, due to time pressures, these flags may not be completed thoroughly. Classification algorithms can be used to check these flags and expand coverage, providing police analysts with a more complete picture of known factors. This kind of classification is particularly beneficial for the scanning stage of POP, as it allows for better grouping of crimes with similar contexts or processes, forming the basis for targeted interventions.

Named Entity Recognition (NER)

In recent years, the importance of combating organized crime related to cross-border criminal activities, such as illegal smuggling, human trafficking, and drug trafficking, has grown significantly.

"This latest report shows how the pandemic has increased vulnerabilities to trafficking in persons, further undercutting capacities to rescue victims and bring criminals to justice"- Ghada Waly, UNODC Executive Director.

Valuable criminal justice analyses based on web texts are often difficult for intelligence investigators to access and use automatically. The manual process of collecting, reading, and analyzing documents from various websites is time-consuming, and the results may not always be satisfactory.

NER can play a crucial role in automating the process of searching for information, extracting desired data, classifying documents, and profiling text authors and organized groups. Identifying and recognizing entities, such as places, organizations, or personal names, can help police and border guard officers understand and find relevant information in the extracted data.


In contrast to classification and named entity recognition, which relies on searching for known characteristics, unsupervised clustering offers the advantage of grouping similar crimes without being constrained by pre-existing administrative categories or preconceived notions of causal factors. This approach allows for more efficient targeting of POP responses.

Clustering can be extended to encompass other variables, such as time and location information, enabling a richer scan for problems than would otherwise be possible. For example, in a burglary case, clustering may provide insights into the emergence of a new modus operandi. In the past, techniques such as hooking keys through letterboxes or snapping certain door locks emerged and were only addressed once they became widespread.

Unsupervised clustering techniques can also prove valuable in the assessment phase of the POP framework. Understanding how criminals are adapting to POP responses is crucial for ensuring lasting impacts from POP interventions. The emergence or shift of crime clusters after a POP evaluation can signal that new techniques are being employed to circumvent the intervention.

By identifying these changes, law enforcement agencies can adapt their strategies accordingly and continue to effectively combat crime. This constant reassessment and adaptation are essential in maintaining a proactive and agile approach to policing.

Addressing Ethical Implications and Biases

While NLP offers numerous opportunities to enhance policing practices, it is crucial to consider the ethical implications and potential biases that may arise when utilizing free-text information from police activities. 

Data Coverage

In many countries, only a fraction of crimes are reported to the police. Factors such as the seriousness of the offense and the level of disadvantage in a neighborhood can influence reporting rates. Consequently, NLP may introduce biases in resource allocation to areas where reporting is more complete, leading to an unfair distribution of resources. To mitigate this issue, law enforcement agencies must account for the underreporting of crime when developing and implementing NLP-based solutions and aim to create equitable strategies that address the needs of underrepresented communities.

Data Richness

The quality of information can affect feedback loops. The quality extracted from free-text data depends on the initial recorded information. Systematic imbalances in the details of recorded crime across areas, communities, or particular groups can result in biases that affect POP responses. To prevent these biases, researchers should examine the richness and overall quality of information recorded across various victim characteristics and crime types. By doing so, they can ensure a more uniform application of POP activities and focus on areas where police-community information flows are efficient.

Algorithmic Bias

Certain crime types disproportionately affect different parts of society, and similar crimes may have similar written descriptions. If specific crime descriptions are not well understood by certain models due to unusual language or other factors, this can result in poorer information retrieval for certain crimes or victim profiles. Like COMPAS (correctional offender management profiling for alternative sanctions), an AI system designed to assist law enforcement in identifying low-risk and high-risk offenders was found to exhibit unintentional bias against African Americans. This issue, known as algorithmic bias, suggests that model applicability should be assessed at a crime-specific level. By reviewing metrics for each crime type, law enforcement agencies can ensure that no crimes or victim types are misrepresented.

Moreover, it is essential to monitor biases in model errors, which may reflect existing recording practices. Doing so will help guarantee that particular crimes and victims are not disadvantaged by specific models.

Pre-trained Language Models in NLP-Powered POP

Pre-trained language models, such as BERT, have gained significant attention due to their success across a range of natural language processing tasks and domains. The adoption of pre-trained language models in the context of problem-oriented policing powered by NLP has significant potential. Here are some of the relevant points to consider:

Reduced Pre-processing Burden

Pre-trained language models are powerful tools that reduce the pre-processing burden for users. By eliminating the need for feature engineering and embedding generation, pre-trained models make it easier for law enforcement agencies to implement NLP-powered POP strategies.

Potential Bias Challenges

However, it is crucial to note that pre-trained models can also perpetuate bias, and the ethical challenges discussed above remain pertinent. Commercial offerings of pre-packaged auto-NLP have the potential to be successful within police agencies, but it is essential to ensure that the models are not creating new or perpetuating existing biases. Users of the system must be able to understand the models or be partnered with an agency that

No items found.


Real-World Journeys

Learn about what we do, who our clients are, and how we create future-ready businesses.
How NLP helps in Digital Marketing?
Natural Language Processing, a branch of Artificial Intelligence, is used by machines to understand human languages. Let's explore this science of communication in depth.
NLP Use Cases in Fintech
Discover the transformative impact of Natural Language Processing (NLP) in Fintech. Explore its varied use cases, from customer service to risk management.
NLP in the Public Sector
Explore how government leverages NLP for service, savings, and engagement

Download Resource

Enter your email to download your requested file.
Thank you! Your submission has been received! Please click on the button below to download the file.
Oops! Something went wrong while submitting the form. Please enter a valid email.