Interoperable Semantic Systems in Public Administration: AI-Driven Data Mining from Law-Enforcement Reports

Abstract

The digitisation of law-enforcement archives is examined with the aim of moving from static analogue records to interoperable semantic information systems. A step-by-step framework for optimal digitisation is proposed, grounded in archival best practice and enriched with artificial-intelligence and semantic-web technologies. Emphasis is placed on semantic data representation, which renders information actionable, searchable, interlinked, and automatically processed. As a proof of concept, a large language model—OpenAI ChatGPT, version o3—was applied to a corpus of narrative police reports, extracting and classifying key entities (metadata, persons, addresses, vehicles, incidents, fingerprints, and inter-entity relationships). The output was converted to Resource Description Framework triples and ingested into a triplestore, demonstrating how unstructured text can be transformed into machine-readable, interoperable data with minimal human intervention. The approach’s challenges—technical complexity, data quality assurance, information-security requirements, and staff training—are analysed alongside the opportunities it affords, such as accelerated access to records, cross-agency interoperability, and advanced analytics for investigative and strategic decision-making. Combining systematic digitisation, AI-driven data extraction, and rigorous semantic modelling ultimately delivers a fully interoperable information environment for law-enforcement agencies, enhancing efficiency, transparency, and evidentiary integrity.

1. Introduction

This study focuses on the digitization process of the archives of law enforcement authorities (LEAs), aiming to utilise the data contained in these archives for the development of semantic information systems. The research question explored is “What is the optimal digitisation process for law enforcement authorities?”.

The study does not merely address the scanning or simple recording of documents in digital form but emphasises the semantic representation of the data included in the archives. Semantic representation seeks to transform data into actionable information, making data searchable and interrelated with other information objects [1,2,3,4,5,6]. The importance of this approach lies in the fact that digitisation should not be limited to a simple analogue reproduction but should promote the dynamic management and utilisation of information [7,8]. The semantic web is recognised as the framework within which this approach can be developed, offering tools for efficient search, data linking, and information retrieval [9]. At the same time, it enables the creation of data networks that can expand the usefulness of information for advanced analysis [9,10,11].

The aim of the study is to analyse the key stages of the digitisation process required for the transition from physical archives to the creation of a semantic information management system. The analysis includes the challenges, the technologies that can be utilised, and the capabilities offered by artificial intelligence for extracting and organising data in a semantic way.

The need for this transition arises from the continuously evolving requirements of LEAs for efficient information management, supporting the more effective handling of incidents [9,12,13,14]. Additionally, the goal is to minimise time-consuming processes, improve data accuracy, and ensure security and confidentiality [12].

The management of archives for law enforcement authorities has a long history, beginning with the first forms of physical recording and document filing [15]. In the early years of operation, the archives consisted mainly of handwritten notes, printed documents, and folders stored in physical archive spaces [15,16,17,18,19,20,21]. These processes, although essential for their time, were characterised by limitations in management, search, and information retrieval.

Over the decades, and with the development of technology, the need for more efficient and organised methods of archival management became evident [22]. The introduction of electronic systems at the end of the 20th century marked an initial attempt at digitisation, using computer systems for recording and storing data [15,23]. However, this transition remained superficial, as most electronic archives retained the form of simple document scans, without possibilities for linking or advanced analysis [16,22].

The need for full integration of digital management began to grow more pressing with the increasing complexity of police data and the demand for faster and more accurate access to information [9,10,19,20,21,22,23]. In this context, agencies started a gradual transition to advanced digital systems, incorporating technologies that allow for the creation, storage, and management of data in a dynamic and efficient manner [17,24,25].

The current challenge lies in the shift from basic digitisation to the semantic representation of data [9]. This process involves not only the conversion of physical records into digital form but also the development of structures that enable the connection, linking, and analysis of information based on semantic standards [9,10,11,24]. In this way, the archives of the LEAs can become more functional, usable, and valuable for meeting modern operational needs.

The originality of this study lies in its attempt to bridge the gap between traditional archival digitisation and semantically enriched information systems, specifically tailored for law enforcement authorities [4,7,10,25,26,27]. Unlike previous works that either focus solely on legal frameworks or on general-purpose digitisation technologies, this study presents a methodological synthesis that combines semantic-web technologies with artificial intelligence for knowledge extraction [9,20,21,28,29,30]. The approach highlights practical applications through real-world inspired use cases and emphasises the potential of semantic representation to enhance interoperability, searchability, and decision-making in security-sensitive domains.

This paper is structured as follows: Section 2 presents the materials and methods used in the study. Section 3 analyses the digitisation process in five consecutive stages, covering the theoretical foundation of knowledge representation, the key components of the semantic web, practical implementation examples, the interaction between ontology and query mechanisms, and finally the role of artificial intelligence in extracting and structuring information. Section 4 provides a comprehensive discussion regarding the implications, limitations, and practical challenges of the proposed approach. Finally, Section 5 concludes the paper, summarising the findings and highlighting future research directions.

2. Materials and Methods

The digital transition of law enforcement authorities is a complex and multidimensional process that involves several successive stages [9]. The process begins with the recording and assessment of the existing archives [16]. In this phase, a detailed inventory of physical records is carried out to determine their preservation status, subject matter, and the priority that should be given to the digitisation of each type of record [31,32]. This documentation is crucial for understanding the volume and complexity of the data to be processed.

The next phase includes the scanning and digitisation of documents. The conversion of physical records into digital form requires the use of high-resolution scanners to ensure the fidelity and accuracy of the captured content. At the same time, the scanned files are stored in appropriate formats that support further processing [31,32].

After scanning, the digitised data are organised and stored in modern electronic databases [9,10,32]. This organisation allows easy access, classification, and retrieval of data by authorised users. The use of specialised systems known as document management systems significantly facilitates this process [9,10,32].

One of the most critical stages of the digital transition is the semantic processing and representation of the data. In this stage, data are enriched with metadata, making them searchable and interrelated. The integration of semantic-web technologies, such as ontologies and semantic networks, enables the connection and deeper understanding of information, adding greater value and functionality to the data [33,34].

Furthermore, data security and confidentiality are prioritised at every stage of the process [35]. Security policies are implemented, encryption is applied, and backups are created to prevent data loss and protect sensitive information [32,34].

Finally, for the successful implementation of the digital transition, training is provided to personnel to familiarise them with the new systems and tools. Ongoing support is necessary to address potential problems and to ensure the efficient operation of the systems [33].

A comprehensive approach to the digital transition does not merely focus on converting data into digital form but aims at creating a dynamic information ecosystem that will enhance the operational capabilities of the law enforcement authorities and improve the overall performance of their data management [9,10].

3. AI-Driven Data Extraction and Semantic-Web Construction

3.1. Knowledge Representation with Ontologies: Data, Information, Knowledge, and Wisdom

Knowledge representation is one of the main objectives of the Semantic Web, as it enables the systematic organisation and visualisation of information [36,37,38,39]. The use of ontologies is a central tool for achieving this goal, allowing data to be understood and connected through clearly defined concepts and relationships [40,41,42,43].

This approach begins with the distinction among data, information, knowledge, and wisdom, known as the DIKW hierarchy (Data, Information, Knowledge, Wisdom) [41,44,45,46]. Data are the basic units of information, such as numbers or text, which by themselves do not carry meaning. When these data are organised and given context, they become information [47]. Information, when analysed and understood, leads to knowledge, which enhances decision-making capabilities. Finally, wisdom emerges from a deeper understanding and the ability to apply knowledge in complex environments [47,48,49].

Ontologies serve as tools for representing these levels of the DIKW hierarchy, providing a structured framework for organising and managing knowledge [36,37,38,39]. Through ontologies, data are enriched with semantic attributes, enabling the linking and understanding of relationships among different pieces of information. For example, in a police data management system, ontologies can define relationships among persons, locations, and incidents, facilitating search and analysis [47,48].

The significance of knowledge representation through ontologies lies in the ability to create semantic networks that enhance interoperability between different systems and facilitate the analysis of complex data [4,5,9,10,47,50,51]. In this way, the law enforcement authorities can fully leverage the capabilities of the semantic web to strengthen their operational capacity and improve the efficiency of information management [9].

3.2. Semantic Web: Classes, Subclasses, Properties, Instances, and Data Information

The structure of the semantic web is based on the concept of organising information into classes and subclasses, while properties and instances are used for further describing and linking data. Classes represent general categories or concepts, such as “Persons,” “Locations,” or “Incidents.” These categories can be further specialised through subclasses, which provide a more detailed description of the data [4,5,9,10,50,52,53,54].

Properties are used to express relationships between classes and subclasses or to describe specific characteristics of instances. For example, in a data management system for law enforcement authorities, a property might be “located in” to connect a location with an incident or “related to” to link a person to a case [9,10,50]. Instances are the specific manifestations of classes or subclasses. For example, in the class “Persons,” an instance could be a specific individual, while in the class “Locations,” an instance could be a particular address [9,10,50]. Data properties are used to store values related to instances, such as identification numbers, dates, or other measurable attributes. The ability to store and manage such data allows for detailed analysis and easier retrieval of information [9,10,50].

This organisational structure offers a clear and comprehensible way of organising information, enhancing interoperability and enabling the use of information across different systems. The adoption of this approach by law enforcement authorities can contribute to improving operational efficiency by allowing faster and more accurate management of complex data [9,10,40,42,47,50].

3.3. Application Examples

To illustrate the application of the semantic web in the context of law enforcement authorities, specific examples are presented:

• Classes and Subclasses:
  • Class: “Incidents.” Subclasses: “Thefts,” “Burglaries,” “Vandalism.”
  • Class: “Persons.” Subclasses: “Suspects,” “Witnesses,” “Police Officers.”
• Properties:
  • Property “involvedΙn”: connecting a person with an incident.
  • Property “locatedΙn”: connecting an incident with a location.
• Instances:
  • For the subclass “Suspects”: Instance: “IoannisPapadopoulos.”
  • For the subclass “Thefts”: Instance: “BicycleΤheftOn_20/10/2024.”
• Data Information:
  • For the instance “IoannisPapadopoulos”:

    i

    Identification Number: “AB123456”

    ii

    Date of Birth: “15/07/1985”

    iii

    Address: “25_Irinis Street,_Thessaloniki.”

  • For the instance “BicycleΤheftOn_20/10/2024”:

    i

    Location: “AristotelousSquare”

    ii

    Description: “Black_bicycle,_brand XYZ.”

    iii

    Case Reference Number: “CASE-20241020.”

• Semantic Linking:
  • The incident “BicycleΤheftOn_20/10/2024” can be connected to the person “IoannisPapadopoulos” through the property “suspectedOf.”
  • The location “AristotelousSquare” can be associated with other incidents, such as “Burglaries” in the same area, providing valuable operational data.

These examples demonstrate the potential of utilising the semantic web to improve data analysis and decision-making for law enforcement authorities while also enhancing interoperability between different information systems.

3.4. The Semantic Web as an Information System: Querying the Ontology

The semantic web operates as a dynamic information system that allows the retrieval and analysis of data through querying ontologies. This capability is based on the use of query languages such as SPARQL (SPARQL Protocol and RDF Query Language), which enables efficient searching and the retrieval of specific information from semantic networks [52,53,55,56].

Through querying, users can extract precise information and connect data belonging to different classes and properties [9,10,52]. For example, in a police data management system, a query can be executed to retrieve all “Theft” incidents that occurred in a specific location or during a particular period of time.

Examples of possible queries include

• Searching for incidents based on location:
  • “Retrieve all incidents that occurred at Aristotelous Square.”
  • SPARQL Query:
  • SELECT ?incident WHERE {
  •     ?incident a :Incident.
  •     ?incident :located_in :Aristotelous_Square.
  • }
• Searching for suspects related to a specific incident:
  • “Which suspects are connected to the bicycle theft on 20/10/2024?”
  • SPARQL Query:
  • SELECT ?suspect WHERE {
  •     :Bicycle_Theft_20241020 :involves ?suspect.
  •     ?suspect a :Suspect.
  • }
• Searching for time patterns in incidents:
  • “Retrieve all incidents that occurred during nighttime hours”.
  • SPARQL Query:
  • SELECT ?incident WHERE {
  •     ?incident a :Incident.
  •     ?incident :time_period ?time.
  •     FILTER (?time >= “22:00”^^xsd:time && ?time <= “06:00”^^xsd:time)
  • }

The use of the semantic web as an information system provides the ability to identify complex relationships among data, enabling the extraction of valuable knowledge. These capabilities can be utilised for decision-making, predicting criminal trends, and improving police operations [52,53,56].

3.5. Use of Artificial-Intelligence Language Models for Data Extraction for Semantic Representation

This subsection presents a focused case study that illustrates how an artificial-intelligence language model can be used to extract structured data from narrative police reports. A small set of authentic-style reports is processed in four stages: first, the model recognises named entities and events; second, the extracted items are organised into interim tabular form that mirrors the core entity types of the target information system; third, the tables are mapped to an OWL ontology to create a semantic representation; and, finally, SPARQL queries are executed over the populated knowledge graph to reveal operational insights. The walk-through highlights the end-to-end workflow—from unstructured text to queryable semantic data—and shows how each step contributes to faster, more reliable information management within law-enforcement environments [57,58,59,60,61,62,63].

The following example has been included to demonstrate, in practical terms, how a language model can transform a narrative police report into structured, machine-readable data. A single incident—recorded through an ordinary control-room call—will be shown first in its original text form and then in the extracted tables used by the information system. Each later step in the subsection builds on this report, illustrating how named entities, locations, dates, and vehicle details are automatically recognised, linked to the ontology, and made available for SPARQL queries.

The text that follows presents a simulated police case consisting of seven distinct incident reports. Each report is written in natural language, reflecting the narrative style typically used by law-enforcement officers when documenting real-world cases. Each textual report is followed by structured tables representing the data extracted by the language model [Table 1, Table 2, Table 3, Table 4, Table 5, Table 6, Table 7, Table 8, Table 9, Table 10, Table 11, Table 12, Table 13, Table 14, Table 15, Table 16, Table 17, Table 18, Table 19, Table 20, Table 21, Table 22, Table 23, Table 24, Table 25, Table 26, Table 27, Table 28, Table 29, Table 30, Table 31, Table 32, Table 33, Table 34, Table 35, Table 36, Table 37, Table 38, Table 39, Table 40 and Table 41]. These are then accompanied by a visual representation that illustrates the semantic representation of the extracted knowledge [Figure 1, Figure 2, Figure 3, Figure 4, Figure 5, Figure 6 and Figure 7]. All the referenced data are available in Supplementary Material File S1 (From Natural Language to the Semantic Web).

Police Control-Room Log—Initial Report (R1)

Date/time received: 7 January 2025, 23:50 h Handler: C/R Operator 4526—Metropolitan Police Service “This is Mr Daniel Carter, of 12 Hazelbrook Road, Brixton, SW2. I have noticed a silver hatchback with registration LC21 FZU parked outside 22 Hazelbrook Road, Brixton for several hours. It has not moved all evening and its hazard lights were flashing earlier. I do not recognise the vehicle and it is causing concern in the neighbourhood.” Action taken: Incident created on CAD; graded “Suspicious Vehicle—SV1”. Local patrol unit S97 informed for area check. PNC enquiry on VRM LC21 FZU requested.

The following tables (Table 1, Table 2, Table 3, Table 4, Table 5 and Table 6) show how the language model transforms the narrative text of the first report (R1) into structured data sets that can be ingested directly by the information system.

Table 1. Presentation of the metadata of Report 1 (R1), as extracted by the OpenAI ChatGPT language model, version o3.

Report ID	Title (Original Heading)	Author/Compiler	Type	Date–Time Recorded
R1	Police Control-Room Log—Initial Report	C/R Operator 4526 (MPS)	Suspicious Vehicle	7 January 2025, 23:50

Table 1. Presentation of the metadata of Report 1 (R1), as extracted by the OpenAI ChatGPT language model, version o3.

Report ID	Title (Original Heading)	Author/Compiler	Type	Date–Time Recorded
R1	Police Control-Room Log—Initial Report	C/R Operator 4526 (MPS)	Suspicious Vehicle	7 January 2025, 23:50

Table 2. Presentation of the persons extracted from Report 1 (R1) by the OpenAI ChatGPT language model, version o3.

Person ID	Surname	First Name	Role
P01	Carter	Daniel	Witness/caller

Table 2. Presentation of the persons extracted from Report 1 (R1) by the OpenAI ChatGPT language model, version o3.

Person ID	Surname	First Name	Role
P01	Carter	Daniel	Witness/caller

Table 3. Presentation of the addresses extracted from Report 1 (R1) by the OpenAI ChatGPT language model, version o3.

Address ID	Street Name	No.	Area/Postcode
A01	Hazelbrook Rd	12	Brixton, SW2
A02	Hazelbrook Rd	22	Brixton, SW2

Table 3. Presentation of the addresses extracted from Report 1 (R1) by the OpenAI ChatGPT language model, version o3.

Address ID	Street Name	No.	Area/Postcode
A01	Hazelbrook Rd	12	Brixton, SW2
A02	Hazelbrook Rd	22	Brixton, SW2

Table 4. Presentation of the vehicles extracted from Report 1 (R1) by the OpenAI ChatGPT language model, version o3.

Vehicle ID	Registration Plate
V01	LC21 FZU

Table 4. Presentation of the vehicles extracted from Report 1 (R1) by the OpenAI ChatGPT language model, version o3.

Vehicle ID	Registration Plate
V01	LC21 FZU

Table 5. Presentation of the incident extracted from Report 1 (R1) by the OpenAI ChatGPT language model, version o3.

Incident ID	Date	Time	Description
I01	7 January 2025	23:50	Suspicious parked vehicle outside 22 Hazelbrook Rd

Table 5. Presentation of the incident extracted from Report 1 (R1) by the OpenAI ChatGPT language model, version o3.

Incident ID	Date	Time	Description
I01	7 January 2025	23:50	Suspicious parked vehicle outside 22 Hazelbrook Rd

Table 6. Presentation of the relationships (entity ↔ entity, property) extracted from Report 1 (R1) by the OpenAI ChatGPT language model, version o3.

Subject (Entity)	Object (Entity)	Property (Relation)
P01	A01	livesIn
P01	I01	reportedIncident
I01	V01	involvesVehicle
I01	A02	hasLocation
R1	I01	documentsIncident
R1	P01	hasReporter
R1	V01	mentionsVehicle
R1	A02	mentionsLocation
I01	P01	calledToPolice

Table 6. Presentation of the relationships (entity ↔ entity, property) extracted from Report 1 (R1) by the OpenAI ChatGPT language model, version o3.

Subject (Entity)	Object (Entity)	Property (Relation)
P01	A01	livesIn
P01	I01	reportedIncident
I01	V01	involvesVehicle
I01	A02	hasLocation
R1	I01	documentsIncident
R1	P01	hasReporter
R1	V01	mentionsVehicle
R1	A02	mentionsLocation
I01	P01	calledToPolice

Figure 1. Semantic representation of the information contained in Report 1 (R1). For this representation, the i2 Analyst’s Notebook information system by IBM was used.

Figure 1. Semantic representation of the information contained in Report 1 (R1). For this representation, the i2 Analyst’s Notebook information system by IBM was used.

Crime Report—Residential Burglary (R2)

Date/time reported: 7 January 2025, 09:00 h Reporting station: Brixton Police Station—front counter Officer taking statement: PC 3184 Roberts Mr Nicholas Peters, resident and owner of 22 Hazelbrook Road, Brixton, SW2, states that an unknown person gained entry to his home between approximately 23:00 h on 6 January and 07:30 h on 7 January 2025. The intruder removed assorted jewellery (estimated value GBP 3200) and an amount of cash (approximately GBP 450) from a bedroom drawer. No signs of forced entry were observed on the front door; rear kitchen window found unlatched. Action taken: Scene-of-crime officers (SOCO) requested for fingerprint and CCTV canvas. Crime reference issued: 05/BRX/0125/23.

The following tables (Table 7, Table 8, Table 9, Table 10 and Table 11) show how the language model transforms the narrative text of the second report (R2) into structured data sets that can be ingested directly by the information system.

Table 7. Presentation of the metadata of Report 2 (R2) by the OpenAI ChatGPT language model, version o3.

Report ID	Title	Author/Compiler	Type	Date–Time Recorded
R2	Crime Report—Residential Burglary	PC 3184 Roberts	Burglar	7 January 2025, 09:00

Table 7. Presentation of the metadata of Report 2 (R2) by the OpenAI ChatGPT language model, version o3.

Report ID	Title	Author/Compiler	Type	Date–Time Recorded
R2	Crime Report—Residential Burglary	PC 3184 Roberts	Burglar	7 January 2025, 09:00

Table 8. Presentation of the persons extracted from Report 2 (R2) by the OpenAI ChatGPT language model, version o3.

Person ID	Surname	First Name	Role
P02	Peters	Nicholas	Victim/owner

Table 8. Presentation of the persons extracted from Report 2 (R2) by the OpenAI ChatGPT language model, version o3.

Person ID	Surname	First Name	Role
P02	Peters	Nicholas	Victim/owner

Table 9. Presentation of the addresses extracted from Report 2 (R2) by the OpenAI ChatGPT language model, version o3.

Address ID	Street Name	No.	Area/Postcode
A02	Hazelbrook Rd	22	Brixton, SW2

Table 9. Presentation of the addresses extracted from Report 2 (R2) by the OpenAI ChatGPT language model, version o3.

Address ID	Street Name	No.	Area/Postcode
A02	Hazelbrook Rd	22	Brixton, SW2

Table 10. Presentation of the incident extracted from Report 2 (R2) by the OpenAI ChatGPT language model, version o3.

Incident ID	Date Range	Description
I02	6 January 2025 23:00 → 7 January 2025 07:30	Residential burglary at 22 Hazelbrook Rd; jewellery (GBP 3200) and cash (GBP 450) stolen

Table 10. Presentation of the incident extracted from Report 2 (R2) by the OpenAI ChatGPT language model, version o3.

Incident ID	Date Range	Description
I02	6 January 2025 23:00 → 7 January 2025 07:30	Residential burglary at 22 Hazelbrook Rd; jewellery (GBP 3200) and cash (GBP 450) stolen

Table 11. Presentation of the relationships (entity ↔ entity, property) extracted from Report 2 (R2) by the OpenAI ChatGPT language model, version o3.

Subject (Entity)	Object (Entity)	Property (Relation)
P02	A02	livesIn
P02	I02	victimOfIncident
I02	A02	hasLocation
R2	P02	hasReporter
R2	I02	documentsIncident
R2	A02	mentionsLocation

Table 11. Presentation of the relationships (entity ↔ entity, property) extracted from Report 2 (R2) by the OpenAI ChatGPT language model, version o3.

Subject (Entity)	Object (Entity)	Property (Relation)
P02	A02	livesIn
P02	I02	victimOfIncident
I02	A02	hasLocation
R2	P02	hasReporter
R2	I02	documentsIncident
R2	A02	mentionsLocation

Figure 2. Semantic representation of the information contained in Report 2 (R2). For this representation, the i2 Analyst’s Notebook information system by IBM was used.

Figure 2. Semantic representation of the information contained in Report 2 (R2). For this representation, the i2 Analyst’s Notebook information system by IBM was used.

Crime Report—Stolen Motor Vehicle (R3)

Date/time reported: 15 June 2024, 14:15 h Reporting station: Camden Town Police Station—front counter Officer taking statement: PC 2741 Hughes Mr Christopher Zacharias, holder of UK driving licence ZACHAR052901, resident at 48 Oakfield Terrace, Camden, NW1, reports the theft of his vehicle, a silver Ford Focus bearing registration LC21 FZU. The vehicle was last seen parked outside his home address at 22:30 h on 14 June 2024. At 07:45 h on 15 June it was discovered missing. No glass fragments or signs of forced entry were present at the scene. Vehicle details: Ford Focus Zetec, 2021 model, VIN ending …6734, fitted with factory alarm (status unknown). Action taken: VRM LC21 FZU circulated on the Police National Computer as “Stolen—Category A”. ANPR hot-list updated; local CCTV enquiries initiated.

The following tables (Table 12, Table 13, Table 14, Table 15, Table 16 and Table 17) show how the language model transforms the narrative text of the third report (R3) into structured data sets that can be ingested directly by the information system.

Table 12. Presentation of the metadata of Report 3 (R3) by the OpenAI ChatGPT language model, version o3.

Report ID	Title	Author/Compiler	Type	Date–Time Recorded
R3	Crime Report—Stolen Motor Vehicle	PC 2741 Hughes	Vehicle theft	15 June 2024, 14:15

Table 12. Presentation of the metadata of Report 3 (R3) by the OpenAI ChatGPT language model, version o3.

Report ID	Title	Author/Compiler	Type	Date–Time Recorded
R3	Crime Report—Stolen Motor Vehicle	PC 2741 Hughes	Vehicle theft	15 June 2024, 14:15

Table 13. Presentation of the persons extracted from Report 3 (R3) by the OpenAI ChatGPT language model, version o3.

Person ID	Surname	First Name	Role
P03	Zacharias	Christopher	Victim/owner

Table 13. Presentation of the persons extracted from Report 3 (R3) by the OpenAI ChatGPT language model, version o3.

Person ID	Surname	First Name	Role
P03	Zacharias	Christopher	Victim/owner

Table 14. Presentation of the addresses extracted from Report 3 (R3) by the OpenAI ChatGPT language model, version o3.

Address ID	Street Name	No.	Area/Postcode
A03	Oakfield Terrace	48	Camden, NW1

Table 14. Presentation of the addresses extracted from Report 3 (R3) by the OpenAI ChatGPT language model, version o3.

Address ID	Street Name	No.	Area/Postcode
A03	Oakfield Terrace	48	Camden, NW1

Table 15. Presentation of the vehicles extracted from Report 3 (R3) by the OpenAI ChatGPT language model, version o3.

Vehicle ID	Registration Plate
V01	LC21 FZU

Table 15. Presentation of the vehicles extracted from Report 3 (R3) by the OpenAI ChatGPT language model, version o3.

Vehicle ID	Registration Plate
V01	LC21 FZU

Table 16. Presentation of the incident extracted from Report 3 (R3) by the OpenAI ChatGPT language model, version o3.

Incident ID	Date Range	Description
I03	14 June 2024 22:30 → 15 June 2024 07:45	Theft of vehicle LC21 FZU from 48 Oakfield Terrace, Camden

Table 16. Presentation of the incident extracted from Report 3 (R3) by the OpenAI ChatGPT language model, version o3.

Incident ID	Date Range	Description
I03	14 June 2024 22:30 → 15 June 2024 07:45	Theft of vehicle LC21 FZU from 48 Oakfield Terrace, Camden

Table 17. Presentation of the relationships (entity ↔ entity, property) extracted from Report 3 (R3) by the OpenAI ChatGPT language model, version o3.

Subject (Entity)	Object (Entity)	Property (Relation)
P03	A03	livesIn
P03	V01	ownsVehicle
P03	I03	victimOfIncident
I03	V01	involvesVehicle
I03	A03	hasLocation
R3	P03	hasReporter
R3	V01	mentionsVehicle
R3	I03	documentsIncident
R3	A03	mentionsLocation

Table 17. Presentation of the relationships (entity ↔ entity, property) extracted from Report 3 (R3) by the OpenAI ChatGPT language model, version o3.

Subject (Entity)	Object (Entity)	Property (Relation)
P03	A03	livesIn
P03	V01	ownsVehicle
P03	I03	victimOfIncident
I03	V01	involvesVehicle
I03	A03	hasLocation
R3	P03	hasReporter
R3	V01	mentionsVehicle
R3	I03	documentsIncident
R3	A03	mentionsLocation

Figure 3. Semantic representation of the information contained in Report 3 (R3). For this representation, the i2 Analyst’s Notebook information system by IBM was used.

Figure 3. Semantic representation of the information contained in Report 3 (R3). For this representation, the i2 Analyst’s Notebook information system by IBM was used.

Recovery Report—Abandoned Vehicle Located (R4)

Date/time located: 2 July 2025, 03:25 h Reporting unit: Response Team B, Southwark Borough—Unit QP14 Attending officers: Sgt 876 McAllister and PC 5628 Dean While conducting routine patrols, officers discovered a silver Ford Focus (registration LC21 FZU) abandoned beneath the railway bridge on 15 Riverbank Way, Bermondsey, SE16. The vehicle matched a Category A stolen marker issued on 15 June 2025 following a report by registered owner Mr Christopher Zacharias (Crime Ref 05/BRX/0125/2). No occupants were present; driver’s door found unlocked, ignition barrel intact, battery disconnected. The area was cordoned for forensic examination. Action taken: PNC record updated to “Stolen—Recovered (pending forensic)”. Scene-of-Crime Officer (SOCO) attendance requested—job ref SOCO/3475/25. Vehicle recovery authorised; lifted to Met impound yard, Charlton.

The following tables (Table 18, Table 19, Table 20, Table 21 and Table 22) show how the language model transforms the narrative text of the fourth report (R4) into structured data sets that can be ingested directly by the information system.

Table 18. Presentation of the metadata of Report 4 (R4) by the OpenAI ChatGPT language model, version o3.

Report ID	Title	Author/Compiler (Unit)	Type	Date–Time Recorded
R4	Recovery Report—Abandoned Vehicle	Response Team B/Unit QP14	Vehicle recovery	2 July 2025, 03:25

Table 18. Presentation of the metadata of Report 4 (R4) by the OpenAI ChatGPT language model, version o3.

Report ID	Title	Author/Compiler (Unit)	Type	Date–Time Recorded
R4	Recovery Report—Abandoned Vehicle	Response Team B/Unit QP14	Vehicle recovery	2 July 2025, 03:25

Table 19. Presentation of the addresses extracted from Report 4 (R4) by the OpenAI ChatGPT language model, version o3.

Address ID	Street Name	No.	Area/Postcode
A04	Riverbank Way	15	Bermondsey, SE16

Table 19. Presentation of the addresses extracted from Report 4 (R4) by the OpenAI ChatGPT language model, version o3.

Address ID	Street Name	No.	Area/Postcode
A04	Riverbank Way	15	Bermondsey, SE16

Table 20. Presentation of the vehicles extracted from Report 4 (R4) by the OpenAI ChatGPT language model, version o3.

Vehicle ID	Registration Plate
V01	LC21 FZU

Table 20. Presentation of the vehicles extracted from Report 4 (R4) by the OpenAI ChatGPT language model, version o3.

Vehicle ID	Registration Plate
V01	LC21 FZU

Table 21. Presentation of the incident extracted from Report 4 (R4) by the OpenAI ChatGPT language model, version o3.

Incident ID	Date	Description
I04	2 July 2025 3:25	Abandoned stolen vehicle LC21 FZU found at 15 Riverbank Way

Table 21. Presentation of the incident extracted from Report 4 (R4) by the OpenAI ChatGPT language model, version o3.

Incident ID	Date	Description
I04	2 July 2025 3:25	Abandoned stolen vehicle LC21 FZU found at 15 Riverbank Way

Table 22. Presentation of the relationships (entity ↔ entity, property) extracted from Report 4 (R4) by the OpenAI ChatGPT language model, version o3.

Subject (Entity)	Object (Entity)	Property (Relation)
I04	V01	involvesVehicle
I04	A04	hasLocation
R4	V01	mentionsVehicle
R4	I04	documentsIncident
R4	A04	MentionsLocation
V01	A04	foundIncident

Table 22. Presentation of the relationships (entity ↔ entity, property) extracted from Report 4 (R4) by the OpenAI ChatGPT language model, version o3.

Subject (Entity)	Object (Entity)	Property (Relation)
I04	V01	involvesVehicle
I04	A04	hasLocation
R4	V01	mentionsVehicle
R4	I04	documentsIncident
R4	A04	MentionsLocation
V01	A04	foundIncident

Figure 4. Semantic representation of the information contained in Report 4 (R4). For this representation, the i2 Analyst’s Notebook information system by IBM was used.

Figure 4. Semantic representation of the information contained in Report 4 (R4). For this representation, the i2 Analyst’s Notebook information system by IBM was used.

Forensic Examination Report—Fingerprint Lift from Recovered Vehicle (R5)

Date/time examined: 2 July 2025, 09:40 h Scene reference: SOCO/3475/25 Examining officers: DC 1458 Malik (Fingerprint Bureau) and SOCO 772 Keane A detailed fingerprint examination was carried out on the silver Ford Focus, registration LC21 FZU, at the Met Impound Yard, Charlton, following recovery earlier the same day (incident R4). Two latent marks of evidential quality were developed and lifted from the interior driver-side door handle: Fingerprint ID FP-001—positively matched via IDENT1 to the registered owner Mr Christopher Zacharias (DOB 29-05-1992, PNC ID ZD-882-01). Fingerprint ID FP-002—high-clarity ridge detail; no current match on IDENT1 or local force databases. Classified as “Unknown Person—Category U”. Action taken: FP-001 recorded as “owner handling”—no further action. FP-002 uploaded to IDENT1 and Interpol AFIS for ongoing search; match alerts enabled. Full fingerprint lift forms (MG 22) and photographic plates appended to case file 05/BRX/0125/25.

The following tables (Table 23, Table 24, Table 25, Table 26, Table 27, Table 28 and Table 29) show how the language model transforms the narrative text of the fifth report (R5) into structured data sets that can be ingested directly by the information system.

Table 23. Presentation of the metadata of Report 5 (R5) by the OpenAI ChatGPT language model, version o3.

Report ID	Title	Author/Compiler (Unit)	Type	Date–Time Recorded
R5	Forensic Examination—Fingerprint Lift (Vehicle)	DC 1458 Malik and SOCO 772 Keane	Forensic examination	2 July 2025 9:40

Table 23. Presentation of the metadata of Report 5 (R5) by the OpenAI ChatGPT language model, version o3.

Report ID	Title	Author/Compiler (Unit)	Type	Date–Time Recorded
R5	Forensic Examination—Fingerprint Lift (Vehicle)	DC 1458 Malik and SOCO 772 Keane	Forensic examination	2 July 2025 9:40

Table 24. Presentation of the persons extracted from Report 5 (R5) by the OpenAI ChatGPT language model, version o3.

Person ID	Surname	First Name	Role/Capacity
P03	Zacharias	Christopher	Registered owner

Table 24. Presentation of the persons extracted from Report 5 (R5) by the OpenAI ChatGPT language model, version o3.

Person ID	Surname	First Name	Role/Capacity
P03	Zacharias	Christopher	Registered owner

Table 25. Presentation of the addresses extracted from Report 5 (R5) by the OpenAI ChatGPT language model, version o3.

Address ID	Facility/Street Name	No.	Area/Postcode
A05	Met Impound Yard, Charlton	–	SE7

Table 25. Presentation of the addresses extracted from Report 5 (R5) by the OpenAI ChatGPT language model, version o3.

Address ID	Facility/Street Name	No.	Area/Postcode
A05	Met Impound Yard, Charlton	–	SE7

Table 26. Presentation of the vehicles extracted from Report 5 (R5) by the OpenAI ChatGPT language model, version o3.

Vehicle ID	Registration Plate
V01	LC21 FZU

Table 26. Presentation of the vehicles extracted from Report 5 (R5) by the OpenAI ChatGPT language model, version o3.

Vehicle ID	Registration Plate
V01	LC21 FZU

Table 27. Presentation of the incident extracted from Report 5 (R5) by the OpenAI ChatGPT language model, version o3.

Table 26—Incident
Incident ID	Date-Time	Description
I05	2 July 2025 09:40	Fingerprint examination of recovered vehicle LC21 FZU

Table 27. Presentation of the incident extracted from Report 5 (R5) by the OpenAI ChatGPT language model, version o3.

Table 26—Incident
Incident ID	Date-Time	Description
I05	2 July 2025 09:40	Fingerprint examination of recovered vehicle LC21 FZU

Table 28. Presentation of the fingerprints extracted from Report 5 (R5) by the OpenAI ChatGPT language model, version o3.

Fingerprint ID	Status/Owner Link
FP-001	Matches owner P03 (IDENT1 confirmed)
FP-002	Unknown person—Category U (no match)

Table 28. Presentation of the fingerprints extracted from Report 5 (R5) by the OpenAI ChatGPT language model, version o3.

Fingerprint ID	Status/Owner Link
FP-001	Matches owner P03 (IDENT1 confirmed)
FP-002	Unknown person—Category U (no match)

Table 29. Presentation of the relationships (entity ↔ entity, property) extracted from Report 5 (R5) by the OpenAI ChatGPT language model, version o3.

Subject (Entity)	Object (Entity)	Property (Relation)
I05	V01	involvesVehicle
I05	A05	hasLocation
FP-001	P03	belongsToPerson
FP-001	I05	foundInIncident
FP-002	I05	foundInIncident
R5	I05	documentsIncident
R5	V01	mentionsVehicle
R5	A05	mentionsLocation
R5	P03	mentionsPerson
R5	FP-001	mentionsFingerprint
R5	FP-002	mentionsFingerprint

Table 29. Presentation of the relationships (entity ↔ entity, property) extracted from Report 5 (R5) by the OpenAI ChatGPT language model, version o3.

Subject (Entity)	Object (Entity)	Property (Relation)
I05	V01	involvesVehicle
I05	A05	hasLocation
FP-001	P03	belongsToPerson
FP-001	I05	foundInIncident
FP-002	I05	foundInIncident
R5	I05	documentsIncident
R5	V01	mentionsVehicle
R5	A05	mentionsLocation
R5	P03	mentionsPerson
R5	FP-001	mentionsFingerprint
R5	FP-002	mentionsFingerprint

Figure 5. Semantic representation of the information contained in Report 5 (R5). For this representation, the i2 Analyst’s Notebook information system by IBM was used.

Figure 5. Semantic representation of the information contained in Report 5 (R5). For this representation, the i2 Analyst’s Notebook information system by IBM was used.

Forensic Examination Report—Fingerprint Lift from Burglary Scene (R6)

Date/time examined: 7 January 2025, 10:00 h Scene reference: SOCO/1124/25 Location: 22 Hazelbrook Road, Brixton, SW2 Examining officers: DC 2197 Patel (Fingerprint Bureau) and SOCO 589 Turner Scene-of-crime officers attended the premises of burglary victim Mr Nicholas Peters (see Crime Ref 05/BRX/0125/25). Two usable latent fingerprints were developed on the inside frame of the rear kitchen window: Fingerprint ID FP-003—matched via IDENT1 to the lawful occupant Mr Nicholas Peters (DOB 12-11-1988, PNC ID PT-441-77). Fingerprint ID FP-002—high-quality ridge detail; identical to the unknown mark lifted from vehicle LC21 FZU on 02 July 2025 (Report R5). No subject currently identified. Action taken: FP-003 logged as “owner handling”—no further action. FP-002 linked to burglary crime file and cross-referenced with vehicle-theft investigation; flag raised for suspect development. Fingerprint imagery and MG 22 forms uploaded to HOLMES and shared with Major Crime Unit.

The following tables (Table 30, Table 31, Table 32, Table 33, Table 34 and Table 35) show how the language model transforms the narrative text of the sixth report (R6) into structured data sets that can be ingested directly by the information system.

Table 30. Presentation of the metadata of Report 6 (R6) by the OpenAI ChatGPT language model, version o3.

Report ID	Title	Author/Compiler (Unit)	Type	Date–Time Recorded
R6	Forensic Examination—Fingerprint Lift (Burglary Scene)	DC 2197 Patel and SOCO 589 Turner	Forensic examination	7 January 2025 10:00

Table 30. Presentation of the metadata of Report 6 (R6) by the OpenAI ChatGPT language model, version o3.

Report ID	Title	Author/Compiler (Unit)	Type	Date–Time Recorded
R6	Forensic Examination—Fingerprint Lift (Burglary Scene)	DC 2197 Patel and SOCO 589 Turner	Forensic examination	7 January 2025 10:00

Table 31. Presentation of the persons extracted from Report 6 (R6) by the OpenAI ChatGPT language model, version o3.

Person ID	Surname	First Name	Role/Capacity
P02	Peters	Nicholas	Victim/lawful occupant

Table 31. Presentation of the persons extracted from Report 6 (R6) by the OpenAI ChatGPT language model, version o3.

Person ID	Surname	First Name	Role/Capacity
P02	Peters	Nicholas	Victim/lawful occupant

Table 32. Presentation of the addresses extracted from Report 6 (R6) by the OpenAI ChatGPT language model, version o3.

Address ID	Street Name	No.	Area/Postcode
A02	Hazelbrook Rd	22	Brixton, SW2

Table 32. Presentation of the addresses extracted from Report 6 (R6) by the OpenAI ChatGPT language model, version o3.

Address ID	Street Name	No.	Area/Postcode
A02	Hazelbrook Rd	22	Brixton, SW2

Table 33. Presentation of the incident extracted from Report 6 (R6) by the OpenAI ChatGPT language model, version o3.

Incident ID	Date-Time	Description
I06	7 January 2025 10:00	Fingerprint examination of burglary scene at 22 Hazelbrook Rd (Crime Ref 05/BRX/0125/25)

Table 33. Presentation of the incident extracted from Report 6 (R6) by the OpenAI ChatGPT language model, version o3.

Incident ID	Date-Time	Description
I06	7 January 2025 10:00	Fingerprint examination of burglary scene at 22 Hazelbrook Rd (Crime Ref 05/BRX/0125/25)

Table 34. Presentation of the fingerprints extracted from Report 6 (R6) by the OpenAI ChatGPT language model, version o3.

Fingerprint ID	Status/Owner Link
FP-002	Unknown person—Category U (previously lifted from vehicle LC21 FZU)
FP-003	Matches victim P02 (IDENT1 confirmed)

Table 34. Presentation of the fingerprints extracted from Report 6 (R6) by the OpenAI ChatGPT language model, version o3.

Fingerprint ID	Status/Owner Link
FP-002	Unknown person—Category U (previously lifted from vehicle LC21 FZU)
FP-003	Matches victim P02 (IDENT1 confirmed)

Table 35. Presentation of the relationships (entity ↔ entity, property) extracted from Report 6 (R6) by the OpenAI ChatGPT language model, version o3.

Subject (Entity)	Object (Entity)	Property (Relation)
I06	A02	hasLocation
I06	FP-003	foundFingerprint
I06	FP-002	foundFingerprint
FP-003	P02	belongsToPerson
R6	I06	documentsIncident
R6	A02	mentionsLocation
R6	P02	mentionsPerson
R6	FP-003	mentionsFingerprint
R6	FP-002	mentionsFingerprint
P02	A02	livesIn
FP-002	A02	foundIncident
FP-003	A02	foundIncident

Table 35. Presentation of the relationships (entity ↔ entity, property) extracted from Report 6 (R6) by the OpenAI ChatGPT language model, version o3.

Subject (Entity)	Object (Entity)	Property (Relation)
I06	A02	hasLocation
I06	FP-003	foundFingerprint
I06	FP-002	foundFingerprint
FP-003	P02	belongsToPerson
R6	I06	documentsIncident
R6	A02	mentionsLocation
R6	P02	mentionsPerson
R6	FP-003	mentionsFingerprint
R6	FP-002	mentionsFingerprint
P02	A02	livesIn
FP-002	A02	foundIncident
FP-003	A02	foundIncident

Figure 6. Semantic representation of the information contained in Report 6 (R4). For this representation, the i2 Analyst’s Notebook information system by IBM was used.

Figure 6. Semantic representation of the information contained in Report 6 (R4). For this representation, the i2 Analyst’s Notebook information system by IBM was used.

Fingerprint Identification Report—Match Notification (R7)

Date/time issued: 8 July 2025, 11:30 h Issuing unit: Metropolitan Police Fingerprint Bureau, Lambeth HQ Officer in charge: Sgt 4010 Gardiner (Senior Fingerprint Examiner) A routine IDENT1 search has produced a positive match for Fingerprint ID FP-002 (previously recorded as “Unknown Person—Category U”, see R5 and R6). The mark corresponds to the right-index impression of Mr Anthony Nikitas (DOB 19-04-1997, PNC ID NK-334-09). Nikitas’s reference set was captured on 5 February 2025 following an arrest in flagrante for attempted burglary at Flat 3, 27 Madeira Street, Islington, N1. The case was discontinued at court owing to insufficient evidence of intent; fingerprints nevertheless remain on file. Confidence level: 12 matching ridge characteristics confirmed by second examiner—conclusive (Grade A). Action recommended: Update burglary case file 05/BRX/0125/25 and vehicle-theft file STOL/LC21FZU/25 to add Nikitas as a suspect. Circulate intelligence report to South Area CID and Vehicle Crime Unit. Consider arrest for questioning under PACE 1984.

The following tables (Table 36, Table 37, Table 38, Table 39, Table 40 and Table 41) show how the language model transforms the narrative text of the seventh report (R7) into structured data sets that can be ingested directly by the information system.

Table 36. Presentation of the metadata of Report 7 (R7) by the OpenAI ChatGPT language model, version o3.

Report ID	Title	Author/Compiler	Type	Date–Time Recorded
R7	Fingerprint Identification—Match Alert	Sgt 4210 Gardiner	Fingerprint identification	8 July 2025 11:30

Table 36. Presentation of the metadata of Report 7 (R7) by the OpenAI ChatGPT language model, version o3.

Report ID	Title	Author/Compiler	Type	Date–Time Recorded
R7	Fingerprint Identification—Match Alert	Sgt 4210 Gardiner	Fingerprint identification	8 July 2025 11:30

Table 37. Presentation of the persons extracted from Report 7 (R7) by the OpenAI ChatGPT language model, version o3.

Person ID	Surname	First Name	Role/Capacity
P04	Nikitas	Anthony	Suspect

Table 37. Presentation of the persons extracted from Report 7 (R7) by the OpenAI ChatGPT language model, version o3.

Person ID	Surname	First Name	Role/Capacity
P04	Nikitas	Anthony	Suspect

Table 38. Presentation of the addresses extracted from Report 7 (R7) by the OpenAI ChatGPT language model, version o3.

Address ID	Street Name	No./Flat	Area/Postcode
A06	Madeira Street	27/Flat 3	Islington, N1

Table 38. Presentation of the addresses extracted from Report 7 (R7) by the OpenAI ChatGPT language model, version o3.

Address ID	Street Name	No./Flat	Area/Postcode
A06	Madeira Street	27/Flat 3	Islington, N1

Table 39. Presentation of the incident extracted from Report 7 (R7) by the OpenAI ChatGPT language model, version o3.

Incident ID	Date-Time	Description
I07	8 July 2025 11:30	IDENT1 match: FP-002 to Anthony Nikitas (Grade A certainty)

Table 39. Presentation of the incident extracted from Report 7 (R7) by the OpenAI ChatGPT language model, version o3.

Incident ID	Date-Time	Description
I07	8 July 2025 11:30	IDENT1 match: FP-002 to Anthony Nikitas (Grade A certainty)

Table 40. Presentation of the fingerprints extracted from Report 7 (R7) by the OpenAI ChatGPT language model, version o3.

Fingerprint ID	Status/Owner Link
FP-002	Now matched to P04 Anthony Nikitas (IDENT1)

Table 40. Presentation of the fingerprints extracted from Report 7 (R7) by the OpenAI ChatGPT language model, version o3.

Fingerprint ID	Status/Owner Link
FP-002	Now matched to P04 Anthony Nikitas (IDENT1)

Table 41. Presentation of the relationships (entity ↔ entity, property) extracted from Report 7 (R7) by the OpenAI ChatGPT language model, version o3.

Subject (Entity)	Object (Entity)	Property (Relation)
FP-002	P04	belongsToPerson
FP-002	I07	matchedInIncident
I07	A06	hasLocation
R7	I07	documentsIncident
R7	P04	mentionsPerson
R7	FP-002	mentionsFingerprint
R7	A06	mentionsLocation
FP-002	A06	foundIncident

Table 41. Presentation of the relationships (entity ↔ entity, property) extracted from Report 7 (R7) by the OpenAI ChatGPT language model, version o3.

Subject (Entity)	Object (Entity)	Property (Relation)
FP-002	P04	belongsToPerson
FP-002	I07	matchedInIncident
I07	A06	hasLocation
R7	I07	documentsIncident
R7	P04	mentionsPerson
R7	FP-002	mentionsFingerprint
R7	A06	mentionsLocation
FP-002	A06	foundIncident

After the tabular extraction step, the language model is instructed to generate OWL-compliant code so that every class, object property, and individual can be stored in a machine-readable knowledge graph. OWL (Web Ontology Language) is a W3C standard that extends RDF with formal semantics, class axioms, and reasoning support. The examples below are shown in Turtle syntax, which is concise and human-readable, but any OWL-compatible tool (e.g., Protégé, Apache Jena) can import the same statements in RDF/XML or JSON-LD.

The core components of the generated ontology include classes representing key entities (e.g., persons, vehicles, incidents), object properties describing their relationships, and datatype properties capturing literal values. An indicative example is presented below:

@prefix le:    <http://example.org/le/>. @prefix rdf:   <http://www.w3.org/1999/02/22-rdf-syntax-ns#>. @prefix owl:   <http://www.w3.org/2002/07/owl#>. @prefix xsd:   <http://www.w3.org/2001/XMLSchema#>. ### --- Classes --- le:Person       rdf:type owl:Class.    # Represents a person involved in a report le:Address      rdf:type owl:Class.    # Residential or incident location le:Vehicle      rdf:type owl:Class.    # Vehicle associated with the case ### --- Object-properties --- le:residesAt    rdf:type owl:ObjectProperty ;    # Person’s residence                           owl:domain le:Person ;                           owl:range  le:Address. le:ownsVehicle rdf:type owl:ObjectProperty ;    # Vehicle ownership                           owl:domain le:Person ;                           owl:range  le:Vehicle.

For extended ontology examples, including datatype properties, individuals extracted from reports, and object property assertions (e.g., fingerprints linked across incidents), see Supplementary Material File S2 (OWL representation).

These OWL triples enable semantic reasoning and querying. For example, the question: “Which incidents involve vehicle V01 and fingerprint FP-002?” can be formulated in SPARQL as follows:

SELECT ?incident   WHERE {    le:V01 le:involvesVehicle ?incident.    le:FP-002 le:foundInIncident ?incident.   }

As illustrated in Figure 1, Figure 2, Figure 3, Figure 4, Figure 5, Figure 6 and Figure 7, the extracted knowledge is semantically represented and machine-processable. The reasoning mechanism is visualised in Figure 8, while the semantic inference leading to perpetrator identification is depicted in Figure 9, leveraging the IBM i2 Analyst’s Notebook system.

Figure 7. Semantic representation of the information contained in Report 7 (R7). For this representation, the i2 Analyst’s Notebook information system by IBM was used.

Figure 7. Semantic representation of the information contained in Report 7 (R7). For this representation, the i2 Analyst’s Notebook information system by IBM was used.

Figure 8. The semantic reasoning process leads to the derivation of secure conclusions. For this representation, the i2 Analyst’s Notebook information system by IBM was used.

Figure 8. The semantic reasoning process leads to the derivation of secure conclusions. For this representation, the i2 Analyst’s Notebook information system by IBM was used.

Figure 9. Semantic representation of the objective of perpetrator identification and crime clarification through the i2 Analyst’s Notebook information system by IBM.

Figure 9. Semantic representation of the objective of perpetrator identification and crime clarification through the i2 Analyst’s Notebook information system by IBM.

4. Discussion

The digitisation process for law enforcement authorities presents a series of challenges and opportunities that must be considered for its successful implementation [18,29,30,64]. The difficulties that arise include the technical complexity of digitisation, the need for significant financial and human resources, as well as issues related to data security and confidentiality [65]. Specifically, the technical complexity is associated with the need to digitise large volumes of data from different sources, which may include handwritten documents, printed materials, and old records.

In the context of data confidentiality and protection, the implementation of semantic technologies must be accompanied by robust cybersecurity and privacy-preserving measures. Data stored in interoperable formats is susceptible to unauthorised access if not properly secured. Therefore, encryption protocols (such as AES-256 for data at rest and TLS for data in transit) are essential. In addition, access control mechanisms—based on role-based access control (RBAC) or attribute-based access control (ABAC)—should be integrated into knowledge management systems to ensure that only authorised personnel can access or modify sensitive records [66,67].

Beyond encryption and access controls, data anonymisation and pseudonymisation techniques are critical to safeguarding personal information, especially when datasets are used for training AI models or shared across agencies [67]. These techniques reduce the risk of re-identification, aligning with GDPR Article 32 (security of processing) and Article 25 (data protection by design and by default), as well as the ethical principle of data minimisation. In high-risk environments, differential privacy can also be applied to ensure that statistical analyses or AI-driven outputs do not compromise individual privacy, even when aggregated data is released [66,67].

To ensure full compliance, adherence to international standards such as ISO/IEC 27001 (information security management) [68], ISO/IEC 27701 (privacy information management) [69], and OWASP [70] guidelines for secure web applications is recommended. These practices not only enhance data security but also help law enforcement authorities demonstrate transparency and accountability, two critical elements in public-sector digitisation initiatives [28,30,66].

Another significant issue is the management of data quality. Many physical records may be damaged or incomplete, making their clear representation in digital form difficult. At the same time, the need to implement high security standards requires the adoption of technologies that ensure the protection of data from unauthorised access or alteration. Furthermore, resistance to change among personnel is another factor that may hinder the transition process [71].

Despite these challenges, digitisation offers significant advantages. One of the most important is the improvement of efficiency and the speed of data access [17,19,25,28,72]. Digital archives allow fast searching and retrieval of information, drastically reducing the time required to perform various tasks. Additionally, digitisation facilitates interoperability between different systems and services, enabling the exchange of information in an effective and secure manner [73].

Compared to traditional methods of case documentation and data handling, semantic digitisation introduces significant enhancements in operational and analytical capabilities. Conventionally, police records are maintained in locked PDF files or handwritten reports, often indexed manually using limited fields—typically just names or dates—without any deeper structure or semantic linkage. These constraints hinder the discovery of connections across cases or between different attributes. For example, suspect profiles, incident locations, and types of offenses may be stored in separate reports without any linking mechanism. This often necessitates ad hoc meetings between senior officers who “know the cases”, relying on memory or experience to identify patterns. However, important details may remain buried within the complexity and volume of raw data [28,30,64].

Moreover, traditional indexing is not resilient to time gaps. As illustrated in our example scenario, crucial information that emerged nearly a year after the initial investigation would likely not be detected under manual search systems. In contrast, the use of ontologies and AI-driven semantic representation facilitates automated reasoning across time-separated data and supports the identification of previously unnoticed links, significantly enhancing the scope of criminal intelligence [29,31,32,33,34].

Another important advantage is the ability to analyse large volumes of data through advanced technologies such as AI language models and the semantic web [20,21,30,64,73,74]. These technologies enable the extraction of valuable knowledge from data, contributing to improved decision-making and operational effectiveness. Finally, digitisation enhances the transparency and reliability of processes, as data are stored and managed in ways that facilitate monitoring and analysis.

To ensure the generalisability and robustness of the proposed framework, its validation on multilingual and multimodal law-enforcement records is essential. This can be achieved by fine-tuning AI language models using police reports and judicial documents from diverse linguistic and structural backgrounds. Such efforts require the coordinated involvement of international organisations like Europol, Eurojust, and Interpol, in collaboration with national departments for international police cooperation. Furthermore, United Nations agencies or region-specific institutions (e.g., Frontex) can provide annotated datasets, ethical oversight, and support for capacity building, ensuring that the system accommodates the specificities of each jurisdiction while remaining interoperable [25,26,27,28].

To address technical complexity, agencies should adopt modular and scalable technologies that can be incrementally integrated into existing infrastructure. Lightweight semantic frameworks and open-source tools can be used to minimise deployment barriers and reduce dependence on proprietary systems. Additionally, phased implementation strategies—beginning with pilot projects—can allow gradual adaptation and validation before full-scale rollout. Resistance to change among personnel can be mitigated through structured training programs tailored to specific roles, emphasising practical benefits and operational improvements. Engaging officers early in the digitisation process fosters ownership and facilitates smoother adoption across departments.

A critical step towards effective adoption is aligning semantic digitisation frameworks with existing law-enforcement information systems [25,26,27,28]. This requires mapping shared knowledge fields—such as name, surname, address, and date of birth—into standardised ontological representations. However, integration poses challenges due to structural inconsistencies in legacy databases. For instance, names may appear in reversed order or encoded using compound fields that mix given names, family names, patronymics, and aliases (e.g., “Abu X” patterns). Similarly, address data are often stored inconsistently—combining street name, building number, city, and postal code in a single field—complicating automated parsing. Addressing such heterogeneity calls for the use of rule-based normalisation procedures and language-specific ontological modules to ensure compatibility without loss of meaning [30,31,32,33,34].

Ensuring cross-border interoperability further amplifies the potential impact of semantic digitisation. Interagency collaboration can be strengthened by harmonising ontological standards across jurisdictions and aligning them with international data exchange protocols. Efforts led by Europol and Interpol, as well as bilateral agreements coordinated by departments of international police cooperation, are crucial for creating unified semantic schemas. These schemas must account for local legal terminologies, linguistic variations, and regional security policies. Developing shared ontological vocabularies and data transformation pipelines can significantly improve real-time information sharing and joint investigative capabilities across national borders.

The balance between challenges and advantages makes the digitisation process a valuable yet demanding task. With the right strategy and the utilisation of appropriate technologies, law enforcement authorities can overcome the difficulties and benefit from the opportunities offered by the digital era.

Ultimately, semantic digitisation serves not only as a technical solution but also as a strategic enabler for transnational cooperation, evidence-based policing, and policy harmonisation across EU and international law-enforcement environments.

Contribution and Comparative Advantage of the Proposed Method

Compared to previous studies, which often address archival digitisation in law enforcement as either a legal-institutional process or a purely technical challenge, this work introduces a methodological framework that integrates AI-driven language models with semantic-web technologies [13,28,31].

While prior approaches typically focused on data storage or metadata tagging [2,29], our method enables the extraction of domain-specific knowledge and its representation through interoperable ontologies that support advanced querying and cross-case inference [3,27,29,30,31,32,33,34,35,36].

Additionally, unlike static digitisation systems, this framework facilitates the dynamic linking of temporally and thematically dispersed entities—a feature particularly critical for criminal intelligence [9,19,27].

The study’s novelty also lies in its application-oriented workflow (Figure 10), which demonstrates how these technologies can be adapted to the high-stakes environment of police operations, while respecting both data confidentiality and operational constraints.

Figure 10. End-to-end semantic digitisation workflow proposed in this study: from unstructured natural language police reports to structured knowledge extraction, ontology creation, and semantic representation using interoperable web standards.

Figure 10. End-to-end semantic digitisation workflow proposed in this study: from unstructured natural language police reports to structured knowledge extraction, ontology creation, and semantic representation using interoperable web standards.

5. Conclusions

This study focuses on investigating the optimal digitisation process for law enforcement authorities, addressing the research question: “What is the optimal digitisation process for law enforcement authorities?” The digitisation of police archives is not merely a technical task but a strategic initiative that requires a combination of advanced technologies and organisational adjustments.

The paper highlights the key challenges involved in managing large volumes of data, ensuring data security and quality, and preparing personnel through appropriate training. At the same time, it emphasises the considerable benefits of digitisation, such as accelerated data access, enhanced interoperability, and the application of artificial-intelligence and semantic-web technologies for analysis and decision-making.

The central role of semantic technologies and AI language models is underscored as a mechanism for transforming unstructured data into actionable knowledge. Through this process, information can be extracted, organised, and semantically represented in interoperable formats, creating a dynamic framework that supports police operations. The proposed pipeline is visually summarised in Figure 10, which illustrates the end-to-end workflow—from raw natural language reports to knowledge extraction, ontology creation, and semantic integration.

Nevertheless, the study is subject to several limitations. Due to confidentiality restrictions, the actual internal workflows and systems of law enforcement authorities could not be disclosed, which reduces the realism of the modelled processes. Consequently, simulated reports were used in place of authentic police documents. In addition, the use of publicly available AI language models (e.g., OpenAI) does not fully reflect the secure, isolated environments required in operational law-enforcement settings. Lastly, the absence of benchmark datasets prevents a rigorous, quantitative evaluation of the extraction performance compared to traditional digitisation methods.

As part of future research directions, it would be valuable to conduct a systematic performance evaluation of different AI models (e.g., BERT, GPT) and ontology engineering tools (e.g., Protégé vs. automated pipelines) using human-annotated reports and standard metrics such as precision, recall, and F1-score. This would provide measurable evidence about the effectiveness and reliability of AI-driven entity extraction under law-enforcement constraints and help determine which tools are best suited for secure and efficient deployment.

Overall, the study concludes that an optimal digitisation strategy requires the structured recording of data, the application of advanced technologies for conversion and semantic organisation, and the integration of interoperable knowledge representations. When supported by the appropriate technological and institutional frameworks, law enforcement authorities can achieve a secure, efficient, and transparent digital transformation.