Large Language Models in Open Government Data Analysis:

A Systematic Mapping Study

Alberto Luciano de Souza Bastos

, Luiz Felipe Cirqueira dos Santos

Shexmo Richarlison Ribeiro dos Santos

, Marcus Vinicius Santana Silva

Marcos Cesar Barbosa dos Santos

, Marcos Venicius Santos

, Marckson F

abio da Silva Santos

Mariano Florencio Mendonc¸a

and Fabio Gomes Rocha

Federal University of Sergipe, S

ao Crist

ao, Sergipe, Brazil

{betobastos.ba, lfcs18ts,shexmor, lowpoc.developer, marcos.cesar.se, Marcksonfabio22,

Keywords:

Large Language Models, Open Government Data, Systematic Mapping Study, Public Administration, Artiﬁcial

Intelligence, Digital Government, Evidence-Based Software Engineering, Government Technology Adoption.

Abstract:

Background: The convergence of Large Language Models (LLMs) and open government data presents

transformative potential for public administration, yet there exists a signiﬁcant gap in understanding adoption

patterns in this emerging domain. Aim: This study analyzes adoption patterns of Large Language Models

in open government data analysis, characterizing researchers’ perceptions about beneﬁts, limitations, and

methodological implications. Method: We conducted a systematic mapping study following Petersen et

al. (2008) guidelines, searching six academic databases. After screening, 24 primary studies were analyzed

covering contribution types, validation methods, government domains, and LLM models. Results: Analysis

revealed GPT model family predominance, with health as priority domain (4 studies), followed by security

and justice (3 studies each). Conversational interfaces and information extraction were dominant functions (9

studies each). Conclusions: The ﬁeld demonstrates evolution toward hybrid solutions integrating LLMs with

structured knowledge resources. Consistent challenges across technologies—ethical issues, privacy concerns,

and data quality—indicate the need for uniﬁed frameworks. Future research should focus on developing

practical solutions to achieve technical maturity comparable to established software engineering ﬁelds.

1 INTRODUCTION

The digital era has transformed government-citizen

relationships through Open Government Data (OGD) -

information produced by government entities for un-

restricted access (Wang et al., 2024). Simultaneously,

Large Language Models (LLMs) emerge as promis-

ing AI technology with remarkable natural language

processing capabilities (Cabral et al., 2024).

The convergence between LLMs and OGD

https://orcid.org/0009-0002-3911-9757

https://orcid.org/0000-0003-4538-5410

https://orcid.org/0000-0003-0287-8055

https://orcid.org/0000-0002-1234-5678

https://orcid.org/0000-0002-7929-3904

https://orcid.org/0009-0006-1645-6127

https://orcid.org/0009-0001-6479-1900

https://orcid.org/0000-0003-0732-3980

https://orcid.org/0000-0002-0512-5406

presents unexplored transformative potential. While

governments digitize public information, much re-

mains in unstructured formats, which limits its prac-

tical use (Siciliani et al., 2024). LLMs can transform

this data into actionable information, enhancing trans-

parency and innovation.

However, a signiﬁcant gap exists in understanding

how LLMs are effectively applied in OGD analysis.

The urgency is justiﬁed by the rapid advancement of

LLMs and their potential for a disruptive impact on

public sector transformation (Androutsopoulou et al.,

2024). Models like GPT-4, Claude and Gemini present

unprecedented capabilities, while ethical and gover-

nance issues emerge as fundamental considerations

(Mureddu et al., 2025; Dua et al., 2025).

This research addresses this gap through systematic

analysis of LLM adoption patterns in OGD contexts,

characterizing researchers’ perceptions about beneﬁts,

limitations, and methodological implications to inform

strategic technology investment decisions and data

404

Bastos, A. L. S., Cirqueira dos Santos, L. F., Ribeiro dos Santos, S. R., Silva, M. V. S., Barbosa dos Santos, M. C., Santos, M. V., Santos, M. F. S., Mendonça, M. F. and Rocha, F. G.

Large Language Models in Open Government Data Analysis: A Systematic Mapping Study.

DOI: 10.5220/0013777300003985

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 21st International Conference on Web Information Systems and Technologies (WEBIST 2025), pages 404-411

ISBN: 978-989-758-772-6; ISSN: 2184-3252

policy implementation in the public sector.

The article is organized as follows: Section 2

presents the theoretical background; Section 3 details

the methodology; Section 4 presents results; Section 5

discusses implications; and Section 6 concludes with

future directions.

2 BACKGROUND

This section provides a theoretical foundation struc-

tured in three parts: Large Language Models funda-

mentals, open government data concepts, and their

intersection in public administration applications.

2.1 Large Language Models:

Fundamentals and Evolution

Large Language Models represent signiﬁcant advances

in natural language processing, demonstrating capa-

bilities in text analysis, content generation, and in-

formation extraction (Bronzini et al., 2024). These

technologies have particular relevance for governmen-

tal applications where automated processing of pub-

lic information becomes critical (Alexopoulos et al.,

2024). Contemporary LLM implementations have

opened possibilities for enhancing public sector efﬁ-

ciency through intelligent data processing and natural

language interfaces (Siciliani et al., 2024).

LLM evolution has been marked by advances in

parameter scale, training data quality, and model archi-

tectures (Donner et al., 2024), enabling sophisticated

applications while questions about reliability and trans-

parency remain crucial (Germani et al., 2024).

2.2 Open Government Data: Concepts

and Challenges

Open Government Data constitutes a fundamental pil-

lar of transparency and innovation in the public sec-

tor (Nikiforova et al., 2024), presenting challenges in

quality, interoperability, and usability (Alexopoulos

et al., 2024). Public data ecosystem evolution has been

driven by standardization and demand for efﬁcient

processing tools (Mureddu et al., 2025), with recent

initiatives focusing on integrating advanced technolo-

gies to improve data accessibility and utility (Siciliani

et al., 2024).

2.3 LLMs in Public Administration

LLM application in governmental contexts has ex-

panded rapidly, covering public policy analysis, citizen

services, and document processing (Androutsopoulou

et al., 2024), demonstrating transformative poten-

tial in modernizing public administration (Sandoval-

Almazan et al., 2024). Practical implementations en-

compass automated document classiﬁcation systems

(Kliimask and Nikiforova, 2024), conversational inter-

faces for public services (Cort

es-Cediel et al., 2023),

and analysis tools for government reports (Pesqueira

et al., 2024), though ethical and legal considerations

remain central to adoption debates (Dua et al., 2025).

Integration with government data shows particular po-

tential in public health trend analysis (Tornimbene

et al., 2025), regulatory document processing (Rizun

et al., 2025), and sustainable development goals moni-

toring (Benjira et al., 2025).

3 RESEARCH METHOD

This study adopts the Systematic Mapping Study

(SMS) approach (Petersen et al., 2008) to organize

and analyze scientiﬁc production about LLM applica-

tions in government open data, allowing identiﬁcation

of trends, gaps, and categorization of contributions.

The methodological structure follows the GQM

(Goal-Question-Metric) model with the objective:

Analyze the adoption of LLMs to characterize their

application from researchers’ perspectives in the con-

text of government open data analysis.

The mapping process followed ﬁve main steps (Pe-

tersen et al., 2008): (1) research questions deﬁnition,

(2) search conduction, (3) screening and selection, (4)

classiﬁcation scheme construction, and (5) data extrac-

tion and mapping, as illustrated in Figure 1.

3.1 Research Questions

Seven exploratory research questions were formulated

(Petersen et al., 2008):

•

RQ1 - What types of research contributions

were produced? This question investigates the

nature of contributions according to established

taxonomies (Shaw, 2003), examining whether stud-

ies focused on theoretical frameworks, empirical

models, practical solutions, or procedural method-

ologies.

•

RQ2 - What validation approaches were em-

ployed in the articles? This question analyzes the

methodological strategies used to validate research

ﬁndings, including evaluation-based assessments,

example-driven demonstrations, experiential evi-

dence, persuasive arguments, and analytical frame-

works.

Large Language Models in Open Government Data Analysis: A Systematic Mapping Study

405

Figure 1: Mapping process, adapted from (Petersen et al., 2008).

•

RQ3 - Which governmental domains were inves-

tigated in the research? This question examines

the speciﬁc public sector areas and governmental

functions where Large Language Model applica-

tions were studied, identifying the scope and focus

of implementation contexts.

•

RQ4 - Which Large Language Model architec-

tures were utilized? This question identiﬁes the

speciﬁc LLM technologies, frameworks, and ar-

chitectural approaches employed in the studies,

examining both proprietary and open-source solu-

tions.

•

RQ5 - What functional roles did LLMs serve

in the research? This question investigates the

primary purposes and applications of LLMs within

governmental contexts, examining their roles in

data processing, analysis, generation, and decision

support systems.

•

RQ6 - Which technical tools and methodologies

were employed? This question analyzes the com-

plementary technologies, platforms, and method-

ological frameworks used alongside LLMs to im-

plement solutions in open government data con-

texts.

•

RQ7 - What types of public datasets were uti-

lized? This question examines the characteristics,

formats, and domains of governmental open data

employed in the studies, investigating data sources

and their structural properties.

3.2 Search Strategy

The search strategy used the PIC model (Kitchenham

et al., 2007): P (Population): Public Data; I (Interven-

tion): LLMs; C (Context): Government. The search

string was:

("Public Data" OR "Open data" OR

"Open information" OR "Transparent

data" OR "Transparent information")

AND ("LLM" OR "Generative AI") AND

("Government" OR "Administration"

OR "Governance" OR "Public

Administration" OR "Public sector"

OR "State")

Searches were conducted in six databases: ACM

Digital Library (559), IEEE Xplore (8), Web of

Science (8), ScienceDirect (317), Scopus (17), and

Springer Link (459). After removing duplicates

(n=34), 1,334 unique articles were screened.

The search of the databases occurred between

March and April 2025.

3.3 Selection Process

Three sequential screening stages were conducted:

1st Reading (title, abstract, keywords) resulted in

86 studies; 2nd Reading (introduction, objectives,

results, conclusions) yielded 43 articles; 3rd Read-

ing (complete analysis) included 24 ﬁnal studies. Ta-

ble 1 presents inclusion/exclusion criteria, and Table 2

shows selection evolution.

Table 1: Inclusion and Exclusion Criteria.

Inclusion Criteria Exclusion Criteria

Articles published between

2017 and 2025.

Duplicate articles identiﬁed

across databases.

Peer-reviewed studies with

full text available.

Works not written in En-

glish.

Studies addressing the appli-

cation of LLMs to open gov-

ernment data.

Partial publications such as

extended abstracts, posters,

or abstracts without com-

plete articles.

Studies that answer, even

partially, one of the deﬁned

research questions.

Studies dealing exclusively

with private data or non-

governmental use of LLMs.

Secondary and tertiary stud-

ies (literature reviews).

Table 2: Evolution of Study Selection.

Stage Excluded Articles Retained Articles

1st Reading (exploratory) 1,282 86

2nd Reading (cross-sectional) 43 43

3rd Reading (analytical) 19 24

Initial Total 1,344 24 ﬁnal articles

Cross-validation between researchers was applied

with complete traceability via Parsifal tool, ensuring

reliability and replicability.

3.4 Data Extraction and Mapping

Categorization used keywording technique (Petersen

et al., 2008) with three facets: Type of Contribution

(Shaw, 2003) (Table 3), Type of Validation (Table 4),

and Application Domain (Table 5). Frequency analysis

generated visualizations of contribution types, tempo-

ral distribution, technologies, and government areas.

WEBIST 2025 - 21st International Conference on Web Information Systems and Technologies

406

Figure 2: Article Distribution by Year, Source and Status.

This study followed validity recommendations (Pe-

tersen et al., 2008) through cross-validation, explicit

criteria, complete traceability via Parsifal, ensuring

transparency and replicability. Data and protocols

are available at: https://github.com/AlbertoBastosMe/

mapping llm ogd.

4 RESULTS

This section presents systematic mapping results

across ﬁve dimensions: study characteristics, method-

ological approaches, application domains, LLM mod-

els, and technological ecosystem.

4.1 Study Selection and Characteristics

The systematic mapping identiﬁed 24 primary studies

meeting the inclusion criteria. Temporal distribution

shows increasing research interest (Figure 2), with

signiﬁcant growth from 2023 onwards, coinciding with

ChatGPT’s widespread adoption.

4.2 Methodological Approaches

Shaw’s taxonomy (Shaw, 2003) reveals qualitative/de-

scriptive models predominance (50.00%, n=12), as

shown in Table 3. Empirical models and proce-

dures/techniques present equivalent representation

(16.67% each, n=4). Speciﬁc solutions constitute

8.33% (n=2), while tools/notations and reports show

the lowest representation (4.17% each, n=1).

Table 3: Distribution of Research Result Types (RQ1).

Result Type Freq. % Articles

Qualitative or descriptive model 12 50.00%

P4; P5; P8; P9; P10; P15; P18; P19;

P20; P22; P23; P24

Empirical model 4 16.67% P2; P3; P6; P12

Procedure or technique 4 16.67% P14; P16; P17; P21

Speciﬁc solution/prototype 2 8.33% P7; P11

Tool or notation 1 4.17% P1

Report 1 4.17% P13

Validation approaches (Table 4) show evaluation as

most prevalent (45.8%, n=11), followed by example-

based validation (25.0%, n=6). Experience and per-

suasion present equivalent representation (12.5% each,

n=3), while analysis represents the least frequent type

(4.2%, n=1).

Table 4: Validation Types by Article (RQ2).

Response Type Freq. % Articles

Evaluation 11 45.8%

P6; P16; P14; P15; P1; P12; P17; P11; P2;

P7; P20

Example 6 25.0% P5; P21; P18; P24; P8; P22

Experience 3 12.5% P10; P3; P13

Persuasion 3 12.5% P19; P4; P9

Analysis 1 4.2% P23

4.3 Application Domains

Analysis reveals concentrated adoption across key gov-

ernmental domains (Table 5). Open data management

dominates (n=5), followed by Health (n=4), and Secu-

rity, Justice, e-Government (n=3 each). This distribu-

tion reﬂects growing maturity in transparency initia-

tives and diversiﬁcation across critical public adminis-

tration sectors.

Table 5: Distribution of primary studies by government do-

main (RQ3).

Government Domain Frequency Studies

Health 4 P3, P4, P13, P20

Security 3 P18, P19, P21

Justice 3 P9, P17, P22

Budget 2 P5, P14

Procurement 1 P7

Open data management 5

P6, P11, P16, P23,

P24

e-Government 3 P2, P10, P15

AI public policies 1 P12

Citizen participation 1 P8

Disinformation 1 P1

Total 24

Data types analysis (Table 6) shows Statistical Data

and Metrics, and Transparency and Accountability

Large Language Models in Open Government Data Analysis: A Systematic Mapping Study

407

Data as most prevalent (n=11 each), evidencing ori-

entation toward government transparency. High fre-

quency of Norms and Legislation (n=10) indicates an

important trend in regulatory text processing.

Table 6: Types of Data Used in Studies (RQ7).

Data Types Frequency Articles

Statistical Data and Metrics 11

P1; P2; P5; P6; P13; P14; P16; P19; P22;

P23; P24

Transparency and Accountability Data 11

P5; P6; P7; P8; P9; P10; P11; P14; P15;

P16; P24

Norms and Legislation 10

P3; P4; P9; P14; P17; P18; P19; P21; P22;

P24

Textual Data and Documents 4 P1; P12; P8; P15

AI and Machine Learning Data 5 P1; P4; P17; P18; P22

Research and Methodology Data 2 P3; P20

Health and Surveillance Data 3 P5; P13; P21

Social Media and Participation Data 3 P8; P15; P13

Metadata and Knowledge Graphs 4 P6; P11; P14; P17

Structured and Semantic Data 3 P7; P12; P16

Geospatial and IoT Data 2 P10; P23

Personal Data and Privacy 3 P9; P13; P21

Multimodal and Emerging Data 3 P18; P10; P13

4.4 LLM Models and Technological

Ecosystem

Clear GPT family predominance emerges (Table 7),

establishing it as the primary governmental choice.

Alternative models like Claude, Llama, and Gemini

appear concentrated in comparative studies (Hannah

et al., 2025).

Table 7: Large Language Models identiﬁed in studies (RQ4).

LLM Frequency Studies

ChatGPT 8

P2; P8; P12; P1; P24; P19;

P4; P13

BERT 6 P1; P16; P14

GPT-4 4 P19; P11; P4; P14

GPT-3.5 2 P19; P11

Gpt 2 P16; P14

GPT-3 1 P4

Ibm Watson Health 1 P5

Rebel 1 P7

Large Language Models 1 P10

Text-Davinci-003 1 P11

Claude 1 P11

T5 1 P11

PaLM 1 P11

Wizardlm 1 P14

Ada 1 P16

Gpt4 Vision 1 P18

Gemini 1 P18

Phi3 1 P18

Sora 1 P19

Bard 1 P24

Functional analysis (Table 8) reveals diverse appli-

cations with information extraction and conversational

interfaces predominating (9 studies each), indicating a

focus on data accessibility and citizen interaction.

Technological ecosystem analysis (Table 9) shows

ChatGPT predominance (3 occurrences), followed by

Hugging Face, BERT, and GPT-4. Five custom tools

were identiﬁed, including TAGIFY (Kliimask and

Nikiforova, 2024), OIE4PA (Siciliani et al., 2024), and

SATIKAS (Sandoval-Almazan et al., 2024), demon-

Table 8: Large Language Model Functions (RQ5).

LLM Function Frequency Studies

Information extraction 9

P5; P11; P1; P14;

P19; P7; P13; P3;

P16

Conversational interface 9

P5; P1; P12; P17; P2;

P18; P24; P8; P20

Content generation 7

P22; P19; P21; P17;

P2; P18; P4

Classiﬁcation 5 P11; P1; P19; P7; P8

Data analysis 4 P5; P3; P17; P6

Summarization 3 P12; P8; P13

NLP 3 P22; P10; P15

Fact-Checking 1 P4

Sentiment analysis 1 P8

NLP in medical records 1 P13

Semantic understanding 1 P14

Structured knowledge generation 1 P14

Information search and retrieval 1 P15

Schema Mapping 1 P16

Multimodal processing 1 P18

Automatic text analysis 1 P20

Analysis support 1 P24

Community facilitator 1 P24

Table 9: Tools and frameworks used in studies with LLMs

(RQ6).

Tool/Framework Frequency Studies

ChatGPT 3 P4; P19; P8

Hugging Face 3 P18; P1; P14

BERT 3 P16; P1; P14

GPT-4 3 P19; P4; P11

Gemini 2 P17; P18

RoBERTa 1 P1

DistilBERT 1 P1

TrustServista API 1 P1

SATIKAS 1 P2

GPT-3 1 P4

IBM Watson Health 1 P5

Python 1 P6

math 1 P6

Pandas 1 P6

regular expressions 1 P6

strating emerging specialization for governmental de-

mands.

5 DISCUSSION

This systematic mapping study provides a compre-

hensive analysis of Large Language Models applica-

tions in open government data contexts, revealing both

signiﬁcant opportunities and critical challenges that

deﬁne the current state of this emerging ﬁeld. The

investigation demonstrates that while LLMs possess

considerable potential for transforming government

data analysis and citizen services, their effective imple-

mentation requires careful consideration of technical,

ethical, and governance factors that span multiple in-

terconnected dimensions of public administration.

WEBIST 2025 - 21st International Conference on Web Information Systems and Technologies

408

5.1

Technological Landscape and Model

Adoption Patterns

The analysis reveals a clear predominance of propri-

etary models in governmental applications, with GPT

family models dominating the landscape (ChatGPT

appearing in 8 studies, GPT-4 in 4 studies, as shown

in Table 7). This concentration reﬂects both acces-

sibility considerations and proven capabilities, but

raises important concerns for public sector implemen-

tations. The reliance on proprietary models creates

potential risks regarding data privacy, vendor lock-

in, and long-term sustainability of governmental AI

initiatives. There exists a signiﬁcant opportunity for

research exploring open-source alternatives such as

Llama, which appeared only in comparative studies

by (Hannah et al., 2025), despite potential advantages

for government use, including enhanced data control,

reduced costs, and greater transparency.

The temporal distribution evidences that research

activity intensiﬁed markedly from 2023 onwards, coin-

ciding with the public release and widespread adoption

of ChatGPT, as shown in Figure 2. This pattern sug-

gests that the ﬁeld’s development has been largely

reactive to commercial technology availability rather

than proactive in addressing speciﬁc governmental re-

quirements.

5.2 Application Domains and Research

Gaps

The domain analysis reveals signiﬁcant imbalances in

research focus that highlight both established strengths

and critical gaps. Health applications represent the

most mature area (4 studies), followed by security

and justice domains (3 studies each), as detailed in

Table 5. However, fundamental governmental areas

remain severely underexplored, particularly education,

which is essential for public service delivery but re-

ceived minimal attention in the analyzed corpus.

The concentration in open data management (5

studies) demonstrates growing recognition of LLMs’

potential for enhancing government transparency ini-

tiatives. Yet, the limited exploration of geospatial data

applications represents a substantial missed opportu-

nity, given the importance of location-based services

and spatial analysis in urban planning, emergency re-

sponse, and infrastructure management.

Furthermore, the geographic and linguistic limita-

tions of current research constrain its global applica-

bility. The predominance of English-language studies

and Western legal frameworks limits the generaliz-

ability of ﬁndings to diverse governmental contexts,

particularly those operating in different languages or

legal traditions. This represents a critical gap, as lin-

guistic and cultural particularities signiﬁcantly impact

LLM performance in governmental applications.

5.3 Methodological Maturity and

Validation Approaches

The distribution of research types according to Shaw’s

taxonomy reveals that 50% of studies employ qualita-

tive or descriptive models (Table 3), indicating the ﬁeld

remains in a theoretical consolidation phase rather than

operational maturity. The limited presence of empiri-

cal models (16.67%) and experimental approaches sug-

gests a critical need for rigorous validation methodolo-

gies that can establish deﬁnitive performance bench-

marks for governmental applications.

The predominance of evaluation-based validation

(45.8% of studies, Table 4) demonstrates methodolog-

ical awareness, yet the scarcity of longitudinal stud-

ies limits understanding of long-term impacts on gov-

ernment efﬁciency and citizen satisfaction. This gap

becomes particularly problematic when considering

the substantial investments required for governmental

AI implementations and the need for evidence-based

decision-making in public sector technology adoption.

5.4 Data Types and Processing

Challenges

The analysis of utilized data types (Table 6) reveals that

Statistical Data and Metrics, along with Transparency

and Accountability Data, dominate applications (11

studies each). While this focus aligns with core govern-

mental transparency objectives, it also highlights the

underutilization of LLMs for processing more com-

plex data types such as multimedia content, citizen

feedback, and real-time sensor data that could enhance

smart city initiatives.

The signiﬁcant representation of Norms and Leg-

islation (10 studies) underscores both the promise

and complexity of applying LLMs to regulatory texts.

However, the concentration on English-language legal

frameworks limits practical applicability to diverse le-

gal systems, particularly those operating under civil

law traditions or incorporating customary legal prac-

tices.

5.5 Research Agenda and Future

Directions

The identiﬁed gaps collectively point toward several

critical research priorities that could advance the ﬁeld

toward operational maturity. First, there is an urgent

Large Language Models in Open Government Data Analysis: A Systematic Mapping Study

409

need for comprehensive comparative studies exam-

ining open-source versus proprietary model perfor-

mance in governmental contexts, particularly address-

ing privacy, security, and cost considerations that are

paramount for public sector implementations.

Second, the development of domain-speciﬁc eval-

uation frameworks for governmental AI applications

represents a fundamental requirement. Unlike general-

purpose AI applications, governmental implementa-

tions must address unique requirements, including

transparency, accountability, fairness, and legal com-

pliance that are not adequately captured by existing

benchmarks.

Third, the expansion of research to underexplored

domains such as education, environmental monitoring,

and citizen engagement platforms could signiﬁcantly

broaden the impact of LLM applications in public ad-

ministration. These areas present substantial opportu-

nities for enhancing service delivery while addressing

critical societal challenges.

Fourth, multilingual and cross-cultural studies are

essential for developing globally applicable govern-

mental AI solutions. Research examining LLM per-

formance across different languages, legal systems,

and cultural contexts would provide crucial insights

for international cooperation and technology transfer

initiatives.

Finally, longitudinal studies examining the sus-

tained impact of LLM implementations on governmen-

tal efﬁciency, citizen satisfaction, and democratic par-

ticipation represent critical knowledge gaps that must

be addressed to inform strategic technology adoption

decisions in the public sector.

The research landscape demonstrates that while

signiﬁcant progress has been made in exploring LLM

applications for open government data, achieving tech-

nical maturity comparable to established software en-

gineering ﬁelds requires coordinated efforts to address

methodological limitations, develop empirical valida-

tion frameworks, and establish ethical governance stan-

dards that can guide responsible innovation in public

sector AI implementations.

6 CONCLUSIONS

This systematic mapping study analyzed LLM adop-

tion patterns in open government data analysis through

a comprehensive examination of 24 primary studies,

providing a consolidated overview of current appli-

cations in governmental contexts and characterizing

researchers’ perceptions about beneﬁts, limitations,

and methodological implications.

Key ﬁndings reveal proprietary model predomi-

nance, particularly GPT family, indicating commercial

solution accessibility while highlighting underexplo-

ration of open-source alternatives better serving gov-

ernmental transparency and data sovereignty require-

ments. Research concentration in health, security, and

open data management demonstrates established ap-

plications, yet signiﬁcant gaps remain in fundamental

services like education and geospatial data processing.

Methodologically, the ﬁeld exhibits theoretical con-

solidation characteristics rather than operational ma-

turity, with 50% employing qualitative/descriptive ap-

proaches, suggesting substantial opportunities for ad-

vancing toward empirical validation and experimen-

tal methodologies establishing deﬁnitive performance

benchmarks.

Despite contributions, several limitations constrain

generalizability. Temporal coverage (2023-2025), 2,

reﬂects the nascent stage but limits the longitudinal per-

spective on technological evolution. English-language

study predominance and Western legal framework fo-

cus constrains applicability to diverse governmental

contexts.

Future research directions emerge from identiﬁed

gaps. Developing standardized evaluation metrics for

LLM performance in government applications repre-

sents a fundamental requirement for evidence-based

practices. Longitudinal studies examining long-term

impacts on transparency, efﬁciency, and citizen engage-

ment constitute critical priorities, while comparative

analyses of open-source versus proprietary models

offer substantial opportunities for advancing public

sector AI governance.

Expanding research to underexplored domains

such as education, environmental monitoring, and mul-

tilingual applications could signiﬁcantly broaden the

transformative potential. Developing domain-speciﬁc

solutions addressing unique governmental require-

ments, including transparency, accountability, and le-

gal compliance, represents an essential pathway to-

ward technical maturity.

While LLM application in open government data

analysis represents a rapidly evolving ﬁeld with trans-

formative potential, achieving operational maturity

requires coordinated efforts addressing methodologi-

cal limitations, developing empirical validation frame-

works, and establishing ethical governance standards.

The comprehensive framework established serves as a

valuable reference for researchers, practitioners, and

policymakers seeking to understand and contribute to

this emerging interdisciplinary domain at the intersec-

tion of artiﬁcial intelligence and digital government.

WEBIST 2025 - 21st International Conference on Web Information Systems and Technologies

410

REFERENCES

Alexopoulos, C., Ali, M., Maratsi, M. I., Rizun, N., Charal-

abidis, Y., Loukis, E., and Saxena, S. (2024). Assessing

the availability and interoperability of open govern-

ment data (ogd) supporting sustainable development

goals (sdgs) and value creation in the gulf cooperation

council (gcc). Quality & Quantity.

Androutsopoulou, M., Askounis, D., Carayannis, E. G., and

Zotas, N. (2024). Leveraging ai for enhanced egovern-

ment: Optimizing the use of open governmental data.

Journal of the Knowledge Economy.

Benjira, W., Atigui, F., Bucher, B., Grim-Yefsah, M., and

Travers, N. (2025). Automated mapping between sdg

indicators and open data: An llm-augmented knowl-

edge graph approach. Data & Knowledge Engineering,

156:102405.

Bronzini, M., Nicolini, C., Lepri, B., Passerini, A., and

Staiano, J. (2024). Glitter or gold? deriving structured

insights from sustainability reports via large language

models. EPJ Data Science, 13:41.

Cabral, B., Souza, M., and Claro, D. B. (2024). Open infor-

mation extraction with llm for the portuguese language.

LINGUAMATICA, 16:167–182.

Cort

es-Cediel, M. E., Segura-Tinoco, A., Cantador, I., and

Rodr

ıguez Bol

ıvar, M. P. (2023). Trends and chal-

lenges of e-government chatbots: Advances in explor-

ing open government data and citizen participation con-

tent. Government Information Quarterly, 40:101877.

Donner, C., Danala, G., Jentner, W., and Ebert, D. (2024).

Truext: Trustworthiness regressor uniﬁed explainable

tool. pages 5325–5334.

Dua, M., Singh, J. P., and Shehu, A. (2025). The ethics of

national artiﬁcial intelligence plans: an empirical lens.

AI and Ethics.

Germani, F., Spitale, G., and Biller-Andorno, N. (2024). The

dual nature of ai in information dissemination: Ethical

considerations. JMIR AI, 3.

Hannah, G., Sousa, R. T., Dasoulas, I., and d’Amato, C.

(2025). On the legal implications of large language

model answers: A prompt engineering approach and a

view beyond by exploiting knowledge graphs. Journal

of Web Semantics, 84:100843.

Kitchenham, B., Budgen, D., and Brereton, P. (2007). Guide-

lines for performing systematic literature reviews in

software engineering. Information and Software Tech-

nology, 49(5–6):481–495.

Kliimask, K. and Nikiforova, A. (2024). Tagify: Llm-

powered tagging interface for improved data ﬁndability

on ogd portals. page 18 – 27. Institute of Electrical and

Electronics Engineers Inc.

Mureddu, F., Paciaroni, A., Pavelka, T., Pemberton, A., and

Remotti, L. A. (2025). Rights and responsibilities: Le-

gal and ethical considerations in adopting local digital

twin technology. pages 291–317.

Nikiforova, A., Lnenicka, M., Mili

c, P., Luterek, M., and

Rodr

ıguez Bol

ıvar, M. P. (2024). From the evolution

of public data ecosystems to the evolving horizons of

the forward-looking intelligent public data ecosystem

empowered by emerging technologies. volume 14841

LNCS, page 402 – 418. Springer Science and Business

Media Deutschland GmbH.

Pesqueira, A., de Bem Machado, A., Bolog, S., Pereira,

R., and Sousa, M. J. (2024). Exploring the impact of

eu tendering operations on future ai governance and

standards in pharmaceuticals. Computers & Industrial

Engineering, 198:110655.

Petersen, K., Feldt, R., Mujtaba, S., and Mattsson, M. (2008).

Systematic mapping studies in software engineering.

International Conference on Evaluation and Assess-

ment in Software Engineering. Available via Research-

Gate.

Rizun, N., Revina, A., and Edelmann, N. (2025). Text

analytics for co-creation in public sector organizations:

a literature review-based research framework. Artiﬁcial

Intelligence Review, 58:125.

Sandoval-Almazan, R., Millan-Vargas, A. O., and Garcia-

Contreras, R. (2024). Examining public managers’

competencies of artiﬁcial intelligence implementation

in local government: A quantitative study. Government

Information Quarterly, 41:101986.

Shaw, M. (2003). Writing good software engineering re-

search papers. In Proceedings of the 25th International

Conference on Software Engineering, ICSE ’03, pages

726–736, Washington, DC, USA. IEEE Computer So-

ciety. Minitutorial.

Siciliani, L., Ghizzota, E., Basile, P., and Lops, P. (2024).

Oie4pa: open information extraction for the public

administration. J. Intell. Inf. Syst., 62(1):273–294.

Tornimbene, B., Rioja, Z. B. L., Brownstein, J., Dunn, A.,

Faye, S., Kong, J., Malou, N., Nordon, C., Rader, B.,

and Morgan, O. (2025). Harnessing the power of ar-

tiﬁcial intelligence for disease-surveillance purposes.

BMC Proceedings, 19:7.

Wang, S., Sun, K., and Zhai, Y. (2024). Dye4ai: Assur-

ing data boundary on generative ai services. page

2281–2295.

Large Language Models in Open Government Data Analysis: A Systematic Mapping Study

411