A Review on the Use of Large Language Models in the Context of

Open Government Data

Daniel Staegemann

, Christian Haertel

, Matthias Pohl

and Klaus Turowski

Magdeburg Research and Competence Cluster VLBA, Otto-von-Guericke University Magdeburg, Magdeburg, Germany

Institute of Data Science, German Aerospace Center (DLR), Jena, Germany

Keywords: Generative AI, GenAI, Large Language Model, LLM, Literature Review, Open Data, Open Government Data.

Abstract: Since ChatGPT was released to the public in 2022, large language models (LLM) have drawn enormous

interest from academia and industry alike. Their ability to create complex texts based on provided inputs

positions them to be a valuable tool in many domains. Moreover, since some time, many governments want

to increase transparency and enable the offering of new services by making their data freely available.

However, these efforts towards Open Government Data (OGD) face various challenges with many being

related to the question how the data can be made easily findable and accessible. To address this issue, the use

of LLMs appears to be a promising solution. To provide an overview of the corresponding research, in this

work, the results of a structured literature review on the use of LLMs in the context of OGD are presented.

Hereby, numerous application areas as well as challenges were identified and described, providing researchers

and practitioners alike with a synoptic overview of the domain.

1 INTRODUCTION

Since ChatGPT was released to the public in 2022,

large language models (LLM) and generative

artificial intelligence (GenAI) have been in the centre

of interest inside and outside of academia (Chang et

al. 2024; Raiaan et al. 2024). Due to their ability to

produce sophisticated outputs based on a provided

prompt, they are widely seen as a promising tool to

enhance the operations of organizations across

numerous domains (Brynjolfsson et al. 2023;

Filippucci et al. 2024; Simons et al. 2024).

While they are generally prone to occasionally

making up information, which is referred to as

hallucinations (Huang et al. 2024; Perković et al.

2024), this can be addressed through techniques such

as specific training and fine-tuning or the utilization

of retrieval augmented generation (RAG). Hereby,

the model is given access to specific data and

documents that it can then draw from to produce

higher quality results that are based on the provided

information (Fan et al. 2024).

https://orcid.org/0000-0001-9957-1003

https://orcid.org/0009-0001-4904-5643

https://orcid.org/0000-0002-6241-7675

https://orcid.org/0000-0002-4388-8914

A huge producer of data are governments.

Whenever services are provided, decisions are made,

or statistics are recorded, this adds to the body of

related data. While this is generally positive, as this

allows for their analysis, this also makes it harder to

manage the resulting data deluge. Further, many

governments have pivoted towards not only keeping

the data and utilizing them themselves, but to also

provide their citizens and the general public with

access to many of these information (Attard et al.

2015; Bonina and Eaton 2020). This Open

Government Data (OGD) movement, in turn,

increases transparency and accountability, might lead

to additional insights and services based on the data,

and can help to increase trust (Janssen et al. 2012;

Kucera and Chlapek 2014).

Yet, the effective use of these data is often rather

challenging, because their sheer volume makes it hard

to get an overview of the available information and

the associated meta-information are often

insufficient, which makes the discovery of potentially

Staegemann, D., Haertel, C., Pohl, M., Turowski and K.

A Review on the Use of Large Language Models in the Context of Open Government Data.

DOI: 10.5220/0013642600003967

In Proceedings of the 14th International Conference on Data Science, Technology and Applications (DATA 2025), pages 651-660

ISBN: 978-989-758-758-0; ISSN: 2184-285X

651

useful data extremely cumbersome (Ahmed 2023;

Ansari et al. 2022; Quarati 2023).

However, combining the high accessibility of

LLMs with the wealth of information contained in

OGD could be a valuable asset for shaping the society

of the future, increasing the involvement of citizens,

and offering a plethora of new services.

Thus, the goal of this work is to explore the

current state of the scientific literature on this

research stream, highlighting the potentials and

challenges and outlining the most promising avenues

for future work. To this end, a structured literature

review (SLR) was conducted. Therefore, within this

paper, the following research question (RQ) shall be

answered:

RQ: What is the current state of the scientific

literature on the use of large language models in the

context of Open Government Data?

To answer the RQ, the remainder of this publication

is structured as follows. Following the introduction,

the SLR is outlined. Then the identified literature is

examined. This is followed by a discussion of the

findings. Finally, a conclusion is given, and avenues

for future work are outlined.

2 THE REVIEW

To answer the RQ, a SLR was conducted. As

frequently pointed out, the value of a SLR highly

depends on its rigour and reproducibility (Kraus et al.

2022; vom Brocke et al. 2009). Therefore, adhering

to common practices (Okoli 2015; vom Brocke et al.

2015), before starting the search, a protocol was

developed to guide the process. In the following, the

corresponding steps, as well as the underlying

considerations and the obtained results are outlined.

To identify the relevant literature, Scopus

was

chosen as the primary source, because it provides a

comprehensive coverage across many scientific

databases and publishers. However, since Scopus

alone usually does not find all relevant papers, as will

become visible in Table 3, multiple other databases

and scientific search engines were used in addition to

ensure a broader coverage. For instance, IEEE

Xplore

(IEEE) was added because of IEEE’s

significance in the computer science domain. These

two were complemented by the AIS electronic

Library

(AISeL), which, inter alia, contains the

proceedings of some of the most renowned

https://www.scopus.com

https://ieeexplore.ieee.org

https://aisel.aisnet.org/

conferences in the information systems domain and

the ACM Digital Library

(ACM), which is operated

by the world’s largest computing society (ACM

History Committee 2025). Finally, Springer Nature

Link

(Springer) was added, since in the past it has

shown to be a strong complement to the

aforementioned sources.

While there are some differences in the design of

the search masks, the search terms used in all of these

were kept as similar as possible. In each case, the

search term consisted of two components.

The first covers the realm of LLMs, while also

including papers that refer to GenAI in general.

Further, due to ChatGPT currently being the most

popular LLM, it was also explicitly included in the

term, whereas others were not. In each case, different

spellings and abbreviations were covered to ensure

comprehensiveness:

Part 1: llm OR "large language model" OR

"generative artificial intelligence" OR "generative

ai" OR "gen ai" OR genai OR gpt OR chatgpt

In the second part, the field of Open Data is

addressed. While the focus of the work is on Open

Government Data, this was done to ensure a broader

coverage and include relevant papers that might have

been missed otherwise. Therefore, the corresponding

term was as follows:

Part 2: “open data” OR “public data”

To make sure that both parts are present in the

found papers, these two parts were connected with an

AND. Further, to increase comprehensiveness, the

terms were not only searched in the title but more

broadly, as shown in Table 1.3

Table 1: The utilization of the search terms.

Source Part 1 used in Part 2 used in

AISeL All Fields All Fields

ACM An

where Title

IEEE All Metadata All Metadata

Scopus Article title,

Abstract, Ke

wor

Article title,

Abstract, Ke

wor

Springe

Keywords Title

Thus, the final search term in Scopus was, for

instance, as follows:

( TITLE-ABS-KEY ( llm OR "large language

model" OR "generative artificial intelligence" OR

"generative ai" OR "gen ai" OR genai OR gpt OR

https://dl.acm.org/

https://link.springer.com//

DATA 2025 - 14th International Conference on Data Science, Technology and Applications

652

chatgpt ) AND TITLE-ABS-KEY ( "Open Data" OR

"public data" ) )

The initial search in Scopus resulted in the

identification of a total of 138 items. IEEE, in turn,

yielded 36 papers and Springer Nature Link 10.

Through ACM, 5 papers were found, and AISeL

contributed 69 additional papers. Thus, overall, the

keyword search brought 258 items. However, since

multiple databases were used for the search, several

duplicates occurred that were removed in the next

step. After doing so, 238 items remained.

Naturally, not each of these fit the intended scope,

which made additional filtering necessary. Aligned

with common practices (vom Brocke et al. 2015), as

shown in Figure 1, this was performed in multiple

steps to assure a high degree of diligence while still

maintaining efficiency.

For all of these phases, a joint set of inclusion and

exclusion criteria, as depicted in Table 2, was defined

in advance to serve as the foundation of the filter

process. Hereby, for a paper to be deemed suitable,

each of the inclusion criteria had to be met, whereas

when at least one of the exclusion criteria applied, it

was removed from the set.

To ensure the necessary quality, it was decided to

only include research articles that were published as

conference papers or journal articles. In turn, other

items such as conference reviews, editorials,

introductions to a minitrack, catch word articles,

comments, or summarizations of panel discussions

were not included. Further, book chapters were also

not considered, since they are usually not peer-

reviewed. This is also the reason why preprint

services like arXiv

were not included in the initial

search, since there are “concerns about the research

accuracy, quality, and credibility of preprints”

(Adarkwah et al. 2024). Due to the provided meta-

data of the publications regarding their type not

always being correct and precise, this required

manual checking. For this reason, the differentiation

between the obtained publication types in the

description of this search process is also only

included after this step. After removing documents of

the wrong type, 202 items remained, with 152 of them

being conference papers and 50 journal articles.

Following this, the language was considered as an

additional factor, to ensure that the authors of the

publication at hand can comprehend the content

without needing the help of translation services,

which might involuntarily distort the content. For this

reason, only papers that were written in English were

kept. This resulted in the removal of three papers,

which were all journal articles. Two of them were

written in Chinese and one in Portuguese.

https://arxiv.org/

Table 2: The search's inclusion and exclusion criteria.

Inclusion Criteria Exclusion Criteria

The paper is written in

lish

The paper is a duplicate

The paper is published

in the

roceedings of a

scientific conference or

in a scientific

ournal

The paper only briefly

mentions Open Govern-

ment Data without actually

discussin

it furthe

The paper focusses on

the application of LLMs

in the context of Open

Government Data

The paper only briefly

mentions LLMs without

actually discussing them

furthe

The paper discusses

application scenarios or

(potential) use cases

The paper is a short paper

(here defined as not having

a length of more than 5

es)

The found item is a

conference review, an

editorial, an introduction

to a minitrack, a catch

word article, a comment,

resents the results of a

anel or is a similar

document type that is no

research article in the

narrower sense

Because the goal was to gain profound insights

into the field, papers that had a length of not more

than five pages were also excluded, since they were

deemed too short to provide the necessary depth.

After this step, 162 papers were left, 116 conference

papers and 46 journal articles.

Once these formal aspects were considered and

the papers were filtered accordingly, the actual

content of the remaining ones was taken into account.

Due to this work’s RQ, the papers had to provide

insights into the use of large language models in the

context of Open Government Data. Thus, items that

focused one aspect and only briefly mentioned the

other one, were not relevant.

In the first step of this phase, the remaining papers

were filtered based on their title. When it was obvious

that a publication did not fit the intended scope, it was

excluded from the list, whereas, in case of uncertainty,

the papers were kept. Once this was finished, 70

conference papers and 18 journal articles were left.

Afterwards, the abstracts and keywords were used

to further filter the list. This, again, reduced the

number of papers significantly, to 26, of which 20

were conference papers and 6 journal articles.

Finally, to concludingly appraise the suitability of

the remaining papers to the RQ’s scope, they were

read in total, and those that did not fit were excluded,

A Review on the Use of Large Language Models in the Context of Open Government Data

653

resulting in a final literature set, as shown in Table 3,

that comprises 16 papers, of which 12 are from

conferences and 4 appeared in journals.

An overview of the identified relevant literature is

given in Table 3. Besides the publication year, the

reference, and the type of paper, it is also shown from

which database the items originate. As can be seen,

for this SLR, ACM and IEEE were not crucial,

whereas the other three each contributed at least one

unique item. However, this could not have been

known in advance and their inclusion still increased

the search’s comprehensiveness.

Figure 1: The search process.

Table 3: The identified papers.

ID Title Yea

pe Found in Reference

Reimagining open data ecosystems: a

practical approach using AI, CI, and

Knowled

e Graphs

2023

Conference

Paper

Scopus

(Ahmed 2023)

ChatGPT Application vis-a-vis Open

Government Data (OGD): Capabilities,

Public Values, Issues and a Research Agenda

2023

Conference

Paper

Scopus

(Loukis et al.

2023)

Can Large Language Models Revolutionalize

Open Government Data Portals? A Case of

Usin

ChatGPT in statistics.

ov.sco

2023

Conference

Paper

Scopus

(Mamalis et al.

2023)

4 BRYT: Automated keyword extraction for

open datasets

2024 Journal

Article

Scopus (Ahmed et al.

2024)

DATA 2025 - 14th International Conference on Data Science, Technology and Applications

654

Table 3: The identified papers(cont).

ID Title Yea

pe Found in Reference

Exploring Interpretability in Open

Government Data with ChatGPT

2024

Conference

Paper

Scopus

(Barcellos et al.

2024)

Instruct Large Language Models for Public

Administration Document Information

Extraction

2024

Conference

Paper

Scopus (Carta et al. 2024)

Aragón Open Data Assistant, Lesson Learned

of an Intelligent Assistant for Open Data

Access

2024

Conference

Paper

Springer

Nature Link

(Del Hoyo-

Alonso et al.

2024)

TAGIFY: LLM-powered Tagging Interface

for Improved Data Findability on OGD portals

2024

Conference

Paper

IEEE;

Scopus

(Kliimask and

ikiforova 2024)

On Enabling Dynamic, Transparent, and

Inclusive Government Consultative Processes

with GenAIOpenGov DSS

2024

Conference

Paper

AISeL

(Marjanovic et al.

2024)

From the Evolution of Public Data

Ecosystems to the Evolving Horizons of the

Forward-Looking Intelligent Public Data

Ecosystem Empowered by Emerging

Technologies

2024

Conference

Paper

Scopus;

Springer

Nature Link

(Nikiforova et al.

2024)

Designing a Large Language Model Based

Open Data Assistant for Effective Use

2024

Conference

Paper

Scopus;

Springer

ature Lin

(Schelhorn et al.

2024)

Unveiling inequality: A deep dive into racial

and gender disparities in US court case

closures

2024

Journal

Article

Scopus

(Takefuji 2024)

Web Open Data to SDG Indicators: Towards

an LLM-Augmented Knowledge Graph

Solution

2025

Conference

Paper

Springer

Nature Link

(Benjira et al.

2025b)

Automated mapping between SDG indicators

and open data: An LLM-augmented

knowledge graph approach

2025

Journal

Article

Scopus

(Benjira et al.

2025a)

The Convergence of Open Data, Linked Data,

Ontologies, and Large Language Models:

Enabling Next-Generation Knowledge

stems

2025

Conference

Paper

Springer

Nature Link

(Cigliano and

Fallucchi 2025)

AI for Good: History, Open Data and Some

ESG-

ased Applications

2025

Journal

Article

Scopus

(O’Leary 2025)

While in total 16 papers were found, one journal

article (Benjira et al. 2025a) is an expanded version

of a conference paper (Benjira et al. 2025b), which is

why the two are grouped in the table. Further,

(Ahmed 2023) and (Ahmed et al. 2024) are also

somewhat related but without one being a direct

expansion of the other one, Therefore, they are

considered as independent publications.

3 FINDINGS

Based on the literature, numerous application avenues

for LLMs in the context of OGD can be identified.

Hereby, a plethora of stakeholders can participate

from their capabilities and the associated added

convenience. These include but are not limited to

citizens, journalists, scientists, companies, and

government entities themselves (Loukis et al. 2023).

However, despite the versatility of tasks that can

be facilitated through LLMs, these can be broadly

divided into two categories, the provisioning and the

utilization of OGD, which will also be reflected in the

following two sub-sections.

3.1 OGD Provisioning

One major stream of research is focused on the

capabilities that LLMs offer in the context of

amending the OGD that are made available with

additional information. As these are currently

A Review on the Use of Large Language Models in the Context of Open Government Data

655

oftentimes rather badly described and lacking in

proper descriptions or keywords to support potential

data consumers in finding datasets that are actually

relevant to them (Ahmed 2023; Ansari et al. 2022;

Quarati 2023), addressing this issue is a significant

step to increase the value obtained from providing the

OGD (Alexopoulos et al. 2024). Hereby, this

becomes increasingly important, the more data are

shared and the higher their complexity is, since

maintaining an overview of them becomes more

challenging.

To address this issue, the most commonly found

utilization is the extraction of suitable keywords to

describe the data and facilitate a more targeted search.

While, theoretically, these could also be manually

determined and provided by the creators themselves,

their current limited availability highlights the

limitations of this approach. Moreover, requiring the

keywords to be submitted alongside the data would

be an additional barrier that could lead to a lowered

willingness to participate in the provisioning of OGD,

which would be counterproductive when trying to

increase participation (Barry and Bannister 2014).

Further, this approach would not cover already

published datasets, which is another downside.

Additionally, maintaining consistency throughout the

assignment of keywords is also not an easy task when

many different government bodies, with potentially

varying intrinsic interest in contributing

(Alexopoulos et al. 2024), are involved. Thus, while

on their own, the keywords could be contentual

accurate, a diverse use of synonyms, spellings, or

interpretations could still turn into a barrier.

Yet, LLMs like ChatGPT appear very promising

for this task, which is why their utilization for

keyword extraction was proposed in several of the

identified papers (Ahmed 2023; Ahmed et al. 2024;

Kliimask and Nikiforova 2024). Additionally, the use

of LLMs to create annotations in a slightly more

general sense was suggested in (Nikiforova et al.

2024). This would reduce the manual effort involved

in the creation of keywords and, at the same time,

allow to facilitate comparability. However, in the

current state, LLMs are not performing flawless in

this task, which might require a certain degree of

human oversight depending on the requirements

regarding the quality of the provided keywords and

annotations (Kliimask and Nikiforova 2024).

Another way to increase the findability of OGD

lies in the identification of themes the datasets belong

to, to allow for a suitable categorization. Again,

similar challenges as with the keyword generation

apply, which could be addressed by utilizing LLMs

https://www.deepl.com

for this task (Ahmed 2023; Ahmed et al. 2024;

Marjanovic et al. 2024).

To further enhance the descriptiveness of datasets,

LLMs can also aide in summarizing their content.

This can either be in the form of information triplets

that are extracted as discussed in (Carta et al. 2024;

Cigliano and Fallucchi 2025) or by providing actual

summaries that allow potential users to understand

the contents of datasets after they have performed a

pre-selection (Marjanovic et al. 2024; Nikiforova et

al. 2024). Depending on the use case, enhancing the

datasets’ interpretability and usability by adding

additional context is another task that LLMs can be

used for as highlighted in (Nikiforova et al. 2024).

Another potential use for LLMs could be the

translation of contents to either support offering one’s

services in multiple languages or to be able to ingest

data from different language sources and integrate

them with one’s own portfolio (Alexopoulos et al.

2024). However, in the example presented in

(Kliimask and Nikiforova 2024), where translations

were incorporated, instead of relying on the

capabilities of ChatGPT, which was used for the

generation of tags, DeepL

was harnessed. Further,

while not explicitly addressed in the identified papers,

data completeness is also a serious issue

(Alexopoulos et al. 2024) that could potentially be

addressed through LLM-based control mechanisms.

Moving from the existing datasets to a more

forward looking perspective, (Loukis et al. 2023)

suggested that LLMs could also be helpful for

analysing the OGD portals’ user requests and usage

data to detect needs as well as to identify data whose

usefulness is limited, which is a somewhat common

issue (Alexopoulos et al. 2024).

3.2 OGD Utilization

While generally the availability of vast OGD sets can

be seen in a positive light, since it shows a desire for

transparency and allows to cover many different

domains and aspects, it also comes with its own

challenges for the (potential) users. Besides the fact

that some of the data might be incomplete or rather

useless for value creation (Alexopoulos et al. 2024),

one of the biggest challenges is the discovery of

datasets or specific information that are relevant to

the respective user’s intentions. While the approaches

mentioned in the previous section can help by

increasing the descriptiveness of the datasets, huge

potential also lies in directly supporting the users with

this task.

For this reason, a lot of the current research is

focused on doing so. However, the diversity of

DATA 2025 - 14th International Conference on Data Science, Technology and Applications

656

proposed approaches is noteworthy, indicating on the

one hand the versatility of LLMs in the context of

OGD but on the other hand also a lack of maturity and

best practices.

As could be expected, based on their strengths,

the harnessing of LLMs to facilitate chatbots or data

assistants that allow users to use natural language to

inquire for information is highly present in the

literature (Del Hoyo-Alonso et al. 2024; Loukis et al.

2023; Nikiforova et al. 2024; Schelhorn et al. 2024).

This way, the data discovery could be considerably

simplified, since users do not need to specifically

search for suitable datasets, which they then have to

explore and analyse, but instead, the just ask for the

information they are interested in. Besides the

improved convenience, this also greatly increases the

accessibility, since the necessary capabilities to

effectively interact with OGD (portals) are not always

a given (Alexopoulos et al. 2024; Schelhorn et al.

2024). Further, while the scope of the studies was

limited, in (Barcellos et al. 2024) and (Schelhorn et al.

2024), the interest of potential users in such a solution

was explored, showing that the incorporation of

LLMs into the interaction with OGD is generally

welcomed.

Besides chatbots that provide the desired answers

by searching across the available datasets, the

identified literature also contains several examples

that focused on the provisioning of information once

the relevant datasets are known. Showing how LLMs

can help in the interaction with databases, (Mamalis

et al. 2023) describes the creation of a LLM-based

chatbot that interacts with the SPARQL endpoint of

the statistical data portal of Scotland. Thus, instead of

having to familiarize themselves with SPARQL

queries, the users can use natural language,

significantly decreasing the barrier for interaction.

Another example can be found in (O’Leary 2025),

where it was experimented with how ChatGPT can

support the interaction with OGD that are available as

Excel-files. Further, in (Cigliano and Fallucchi 2025),

the use of LLMs as natural language interfaces to

extract information from knowledge graphs is

proposed. While the authors’ suggestion is made

without mentioning a specific application domain,

this could help in making certain OGD more

accessible to laypeople.

Moreover, as also mentioned in the section

focused on the OGD provisioning, LLMs ability to

provide summaries of information that would

otherwise need a long time to be explored can be

highly valuable, especially when it comes to

developing an initial understanding of a topic (Loukis

et al. 2023; Marjanovic et al. 2024; Nikiforova et al.

2024).

Another aspect where LLMs might be of help is

the linking of information across different datasets to

provide more sophisticated insights (Benjira et al.

2025a, 2025b; Cigliano and Fallucchi 2025). Its

utility can, for instance, be highlighted on the

example of tracking the fulfilment of sustainable

development goals, which are given in one document,

based on corresponding indicators that are spread

across a plethora of datasets (Benjira et al. 2025a,

2025b). To limit the manual work required for the

matching based on the context, LLMs could be

harnessed.

For concepts, where users can either submit

requests for the provisioning of data or suggests

certain government measures, LLMs can help in

several ways. On the one hand, the aforementioned

creation of summaries can allow users to quickly get

an idea of already existing propositions to avoid

duplicate request. On the other hand, LLMs can assist

the users in writing their own proposals by helping

them to adhere to guidelines regarding aspects such

as language, writing style, or structure (Marjanovic et

al. 2024). This reduces the required effort for users to

provide suitable proposals, reducing the barrier for

doing so, while also reducing the effort on the part of

the operators of the portal because the (formal)

quality of the obtained requests is increased and their

structure is better aligned with internal standards and

formatting specifications, simplifying their

processing.

Another potential use case that is mentioned in

(Loukis et al. 2023) and exemplarily demonstrated in

(Takefuji 2024), is the harnessing of LLMs to provide

assistance to programmers for creating OGD-based

applications. This way, utilizing OGD to generate

economic or social value from the available datasets

becomes more accessible, allowing a bigger audience

to contribute. Thereby, on a basic level, LLMs could

be seen as an even more accessible alternative to the

low-code development concept (Hintsch et al. 2021).

4 DISCUSSION

As the analysis of the identified literature showed,

LLMs can be harnessed to perform or aid with a

plethora of tasks related to OGD (Nikiforova et al.

2024). Hereby, their main potential is currently seen

in making the available data more accessible and

easier to handle. Instead of being mostly limited to

users that possess the technical skills and the required

OGD-literacy to effectively navigate the data

A Review on the Use of Large Language Models in the Context of Open Government Data

657

(Alexopoulos et al. 2024), LLMs can this way, a

broader audience is enabled to interact with them,

which, in turn, increases their provided benefit and

strengthens the citizens’ engagement and

participation in the government.

Further, the use of LLMs can also facilitate the

improvement of existing government-related services

as well as the creation of new ones (Loukis et al.

2023). Hereby, these can be entirely based on OGD

or by combining OGD and private data. An example

for such a service is discussed in (Costa et al. 2024)

where data that indicate bike-accidents are combined

with map data to create an easily understandable

visual map indicating different risk-zones for cyclists

in the city of Porto. However, since in this example,

the origin (are they derived from OGD or, for instance,

from private hospitals or services) of the accident data

is not entirely clear, this paper was not included in the

literature review. Yet, it clearly shows the potential

and is therefore mentioned in this place.

Nevertheless, despite the great potential, the use

of LLMs in the context of OGD also comes with

significant challenges that need to be thoroughly

considered. This is also reflected in the literature,

where multiple authors highlight that (at least current

generation) LLMs are still sometimes producing

erroneous responses (Cigliano and Fallucchi 2025;

Del Hoyo-Alonso et al. 2024; Kliimask and

Nikiforova 2024; Loukis et al. 2023; Schelhorn et al.

2024). Consequently, improving the models’

accuracy and consistency or incorporating further

control and correction measures is important for those

cases, where a high quality of the results is essential.

This might, for instance, be especially relevant in

cases where the data are used as a basis for

discussions on very controversial topics, when

mistakes by the LLM could be misinterpreted as

deliberate falsification of facts, eroding the trust in the

data, the portal, or even the government itself.

Further, these models are usually black box in

nature, making it hard to verify and understand the

results, which could deter users from utilizing them

(Loukis et al. 2023). This, however, could be

addressed by integrating explainable AI principles to

increase trust (Nikiforova et al. 2024; Schelhorn et al.

2024; Zoeten et al. 2024).

Other areas that were pointed out in the literature

as potential barriers are related to data and user

privacy, ethical concerns and regulatory frameworks

(Loukis et al. 2023; Nikiforova et al. 2024), whereby

especially the latter, however, also applies to OGD in

general (Alexopoulos et al. 2024).

It is mentioned that linguistic differences across

countries could result in algorithmic biases that might

compromise the performance of the used LLMs

(Loukis et al. 2023).

Moreover, OGD consumers are sometimes faced

with incomplete or erroneous data as well as with data

formats that are highly heterogeneous, incompatible,

or inappropriate (Alexopoulos et al. 2024). Yet, while

LLMs can help in increasing the quality of data and

also in integrating data from different sets and in

different formats with each other, they also require

high-quality, and if possible, well-curated input data

in order to produce reliable and meaningful outputs

(Nikiforova et al. 2024).

5 CONCLUSION

While the opening of government data to the public

can bring enormous benefits through increased

transparency and the potential creation of new

services, accessing and analysing these data can be

quite challenging. However, with the emergence of

LLMs, a powerful new tool has become available that

can facilitate the provisioning and utilization of OGD

in many ways. In the publication at hand, a structured

literature review is described that was conducted to

compile the scientific literature on the use of LLMs

in the context of OGD. Hereby, numerous application

areas were identified and described. Further, in

addition to the opportunities, also potential challenges

were outlined. This way, researchers and practitioners

alike are provided with an overview that allows them

to develop an understanding of the domain, which

they can then use to inform their own work. However,

this work can only be seen as an introduction. To gain

a deeper understanding, especially with regard to the

technical details of the described implementations,

the reader is advised to consult the identified papers.

Further, the study’s focus was only on OGD. Yet,

expanding it to the use of LLMs in the context of open

data in general may yield additional insights that

might also be applicable in the realm of OGD.

Therefore, this appears to be a worthwhile direction

for future research. Overall, the conducted research

showed that there is a lot of activity in the field,

highlighting the perceived potential of LLMs to

facilitate the use of OGD. Though, it also became

clear that the field is still rather immature as indicated

by the many different approaches and strategies being

explored due to a lack of established best practices.

DATA 2025 - 14th International Conference on Data Science, Technology and Applications

658

REFERENCES

ACM History Committee. (2025). “ACM History,”

available at https://www.acm.org/about-acm/acm-

history, accessed on Mar 14 2025.

Adarkwah, M. A., Islam, A. Y. M. A., Schneider, K.,

Luckin, R., Thomas, M., and Spector, J. M. (2024).

“Are Preprints a Threat to the Credibility and Quality

of Artificial Intelligence Literature in the ChatGPT

Era? A Scoping Review and Qualitative Study,”

International Journal of Human–Computer Interaction,

pp. 1-14 (doi: 10.1080/10447318.2024.2364140).

Ahmed, U. (2023). “Reimagining open data ecosystems: a

practical approach using AI, CI, and knowledge graphs,”

in BIR Workshops, Ascoli Piceno, Italy. 13.09.2023 -

15.09.2023.

Ahmed, U., Alexopoulos, C., Piangerelli, M., and Polini, A.

(2024). “BRYT: Automated keyword extraction for

open datasets,” Intelligent Systems with Applications

(23), p. 200421 (doi: 10.1016/j.iswa.2024.200421).

Alexopoulos, C., Saxena, S., Janssen, M., Rizun, N.,

Lnenicka, M., and Matheus, R. (2024). “Why do Open

Government Data initiatives fail in developing

countries? A root cause analysis of the most prevalent

barriers and problems,” The Electronic Journal of

Information Systems in Developing Countries (90:2)

(doi: 10.1002/isd2.12297).

Ansari, B., Barati, M., and Martin, E. G. (2022).

“Enhancing the usability and usefulness of open

government data: A comprehensive review of the state

of open government data visualization research,”

Government Information Quarterly (39:1), p. 101657

(doi: 10.1016/j.giq.2021.101657).

Attard, J., Orlandi, F., Scerri, S., and Auer, S. (2015). “A

systematic review of open government data initiatives,”

Government Information Quarterly (32:4), pp. 399-418

(doi: 10.1016/j.giq.2015.07.006).

Barcellos, R., Bernardini, F., Zuiderwijk, A., and Viterbo,

J. (2024). “Exploring Interpretability in Open

Government Data with ChatGPT,” in Proceedings of

the 25th Annual International Conference on Digital

Government Research, H.-C. Liao, D. D. Cid, M. A.

Macadar and F. Bernardini (eds.), Taipei Taiwan.

11.06.2024 - 14.06.2024, New York, NY, USA: ACM,

pp. 186-195 (doi: 10.1145/3657054.3657079).

Barry, E., and Bannister, F. (2014). “Barriers to open data

release: A view from the top,” Information Polity

(19:1,2), pp. 129-152 (doi: 10.3233/IP-140327).

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

Travers, N. (2025a). “Automated mapping between

SDG indicators and open data: An LLM-augmented

knowledge graph approach,” Data & Knowledge

Engineering (156), p. 102405 (doi: 10.1016/j.

datak.2024.102405).

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

Travers, N. (2025b). “Web Open Data to SDG

Indicators: Towards an LLM-Augmented Knowledge

Graph Solution,” in Web Information Systems

Engineering – WISE 2024 PhD Symposium, Demos and

Workshops, M. Barhamgi, H. Wang, X. Wang, E.

Aïmeur, M. Mrissa, B. Chikhaoui, K. Boukadi, R. Grati

and Z. Maamar (eds.), Springer Nature Singapore, pp.

90-100 (doi: 10.1007/978-981-96-1483-7_7).

Bonina, C., and Eaton, B. (2020). “Cultivating open

government data platform ecosystems through

governance: Lessons from Buenos Aires, Mexico City

and Montevideo,” Government Information Quarterly

(37:3), p. 101479 (doi: 10.1016/j.giq.2020.101479).

Brynjolfsson, E., Li, D., and Raymond, L. (2023).

“Generative AI at Work,”

NBER Working Paper Series

31161, Cambridge, MA: National Bureau of Economic.

Carta, S., Giuliani, A., Manca, M. M., Piano, L., Pisu, A.,

and Tiddia, S. G. (2024). “Instruct Large Language

Models for Public Administration Document

Information Extraction,” in Proceedings of the Ital-IA

2024 Thematic Workshops, Naples, Italy. 29.05.2024 -

30.05.2024, pp. 424-429.

Chang, Y., Wang, X., Wang, J., Wu, Y., Yang, L., Zhu, K.,

Chen, H., Yi, X., Wang, C., Wang, Y., Ye, W., Zhang,

Y., Chang, Y., Yu, P. S., Yang, Q., and Xie, X. (2024).

“A Survey on Evaluation of Large Language Models,”

ACM Transactions on Intelligent Systems and

Technology (15:3), pp. 1-45 (doi: 10.1145/3641289).

Cigliano, A., and Fallucchi, F. (2025). “The Convergence

of Open Data, Linked Data, Ontologies, and Large

Language Models: Enabling Next-Generation

Knowledge Systems,” in Metadata and Semantic

Research, M. Sfakakis, E. Garoufallou, M. Damigos, A.

Salaba and C. Papatheodorou (eds.), pp. 197-213 (doi:

10.1007/978-3-031-81974-2_17).

Costa, D. G., Silva, I., Medeiros, M., Bittencourt, J. C. N.,

and Andrade, M. (2024). “A method to promote safe

cycling powered by large language models and AI

agents,” MethodsX (13), p. 102880 (doi:

10.1016/j.mex.2024.102880).

Del Hoyo-Alonso, R., Rodrigalvarez-Chamarro, V., Vea-

Murgía, J., Zubizarreta, I., and Moyano-Collado, J.

(2024). “Aragón Open Data Assistant, Lesson Learned

of an Intelligent Assistant for Open Data Access,” in

Chatbot Research and Design, A. Følstad, T. Araujo, S.

Papadopoulos, E. L.-C. Law, E. Luger, M. Goodwin, S.

Hobert and P. B. Brandtzaeg (eds.), Cham: Springer

Nature, pp. 42-57 (doi: 10.1007/978-3-031-54975-5_3).

Fan, W., Ding, Y., Ning, L., Wang, S., Li, H., Yin, D., Chua,

T.-S., and Li, Q. (2024). “A Survey on RAG Meeting

LLMs: Towards Retrieval-Augmented Large Language

Models,” in Proceedings of the 30th ACM SIGKDD

Conference on Knowledge Discovery and Data Mining,

R. Baeza-Yates and F. Bonchi (eds.), Barcelona Spain.

25.08.2024 - 29.08.2024, New York, NY, USA: ACM,

pp. 6491-6501 (doi: 10.1145/3637528.3671470).

Filippucci, F., Gal, P., Jona-Lasinio, C., Leandro, A., and

Nicoletti, G. (2024). “The impact of Artificial

Intelligence on productivity, distribution and growth:

Key mechanisms, initial evidence and policy

challenges,” OECD Artificial Intelligence Papers.

Hintsch, J., Staegemann, D., Volk, M., and Turowski, K.

(2021). “Low-code Development Platform Usage:

Towards Bringing Citizen Development and Enterprise

IT into Harmony,” in Proceedings of the 32nd

A Review on the Use of Large Language Models in the Context of Open Government Data

659

Australasian Conference on Information Systems

(ACIS2021), Online. 06.12.2021 - 10.12.2021.

Huang, L., Yu, W., Ma, W., Zhong, W., Feng, Z., Wang, H.,

Chen, Q., Peng, W., Feng, X., Qin, B., and Liu, T.

(2024). “A Survey on Hallucination in Large Language

Models: Principles, Taxonomy, Challenges, and Open

Questions,” ACM Transactions on Information Systems

(doi: 10.1145/3703155).

Janssen, M., Charalabidis, Y., and Zuiderwijk, A. (2012).

“Benefits, Adoption Barriers and Myths of Open Data

and Open Government,” Information Systems

Management (29:4), pp. 258-268 (doi:

10.1080/10580530.2012.716740).

Kliimask, K., and Nikiforova, A. (2024). “TAGIFY: LLM-

powered Tagging Interface for Improved Data

Findability on OGD portals,” in 2024 Fifth

International Conference on Intelligent Data Science

Technologies and Applications (IDSTA), Dubrovnik,

Croatia. 24.09.2024 - 27.09.2024, IEEE, pp. 18-27 (doi:

10.1109/IDSTA62194.2024.10746941).

Kraus, S., Breier, M., Lim, W. M., Dabić, M., Kumar, S.,

Kanbach, D., Mukherjee, D., Corvello, V., Piñeiro-

Chousa, J., Liguori, E., Palacios-Marqués, D.,

Schiavone, F., Ferraris, A., Fernandes, C., and Ferreira,

J. J. (2022). “Literature reviews as independent studies:

guidelines for academic practice,” Review of

Managerial Science (16:8), pp. 2577-2595 (doi:

10.1007/s11846-022-00588-8).

Kucera, J., and Chlapek, D. (2014). “Benefits and Risks of

Open Government Data,” Journal of Systems

Integration, pp. 30-41 (doi: 10.20470/jsi.v5i1.185).

Loukis, E., Saxena, S., Rizun, N., Maratsi, M. I., Ali, M.,

and Alexopoulos, C. (2023). “ChatGPT Application vis-

a-vis Open Government Data (OGD): Capabilities,

Public Values, Issues and a Research Agenda,” in Elec-

tronic Government, I. Lindgren, C. Csáki, E. Kalampokis,

M. Janssen, G. Viale Pereira, S. Virkar, E. Tambouris and

A. Zuiderwijk (eds.), Springer Nature Switzerland, pp.

95-110 (doi: 10.1007/978-3-031-41138-0_7).

Mamalis, M. E., Kalampokis, E., Karamanou, A., Brimos, P.,

and Tarabanis, K. (2023). “Can Large Language Models

Revolutionalize Open Government Data Portals? A Case

of Using ChatGPT in statistics.gov.scot,” in Proceedings

of the 27th Pan-Hellenic Conference on Progress in

Computing and Informatics, N. N. Karanikolas, M. G.

Vassilakopoulos, C. Marinagi, A. Kakarountas and I.

Voyiatzis (eds.), Lamia, Greece. 24.11.2023 -

26.11.2023, New York, NY, USA: ACM, pp. 53-59 (doi:

10.1145/3635 059.3635068).

Marjanovic, O., Hristova, D., Dash, S., and and Abedin, B.

(2024). “On Enabling Dynamic, Transparent, and

Inclusive Government Consultative Processes with

GenAIOpenGov DSS,” in ACIS 2024 Proceedings,

Canberra, Australia. 04.12.2024 - 06.12.2024.

Nikiforova, A., Lnenicka, M., Milić, P., Luterek, M., and

Rodríguez Bolívar, M. P. (2024). “From the Evolution

of Public Data Ecosystems to the Evolving Horizons of

the Forward-Looking Intelligent Public Data Eco-

system Empowered by Emerging Technologies,” in

Electronic Government, M. Janssen, J. Crompvoets, J.

R. Gil-Garcia, H. Lee, I. Lindgren, A. Nikiforova and

G. Viale Pereira (eds.), Springer Nature Switzerland, pp.

402-418 (doi: 10.1007/978-3-031-70274-7_25).

O’Leary, D. (2025). “AI for Good: History, Open Data and

Some ESG-based Applications,” Journal of Decision

Systems (34:1) (doi: 10.1080/12460125.2024.2443182).

Okoli, C. (2015). “A Guide to Conducting a Standalone

Systematic Literature Review,” Communications of the

Association for Information Systems (37), pp. 879-910

(doi: 10.17705/1CAIS.03743).

Perković, G., Drobnjak, A., and Botički, I. (2024).

“Hallucinations in LLMs: Understanding and

Addressing Challenges,” in 2024 47th MIPRO ICT and

Electronics Convention (MIPRO), Opatija, Croatia.

20.05.2024 - 24.05.2024, IEEE, pp. 2084-2088 (doi:

10.1109/MIPRO60963.2024.10569238).

Quarati, A. (2023). “Open Government Data: Usage trends

and metadata quality,” Journal of Information Science

(49:4), pp. 887-910 (doi: 10.1177/016555152 11027775).

Raiaan, M. A. K., Mukta, M. S. H., Fatema, K., Fahad, N. M.,

Sakib, S., Mim, M. M. J., Ahmad, J., Ali, M. E., and

Azam, S. (2024). “A Review on Large Language

Models: Architectures, Applications, Taxonomies, Open

Issues and Challenges,” IEEE Access (12), pp. 26839-

26874 (doi: 10.1109/ACCESS.2024.3365742).

Schelhorn, T. C., Gnewuch, U., and Maedche, A. (2024).

“Designing a Large Language Model Based Open Data

Assistant for Effective Use,” in Design Science Research

for a Resilient Future, M. Mandviwalla, M. Söllner and

T. Tuunanen (eds.), Springer Nature Switzerland, pp.

398-411 (doi: 10.1007/978-3-031-61175-9_27).

Simons, W., Turrini, A., and Vivian, L. (2024). “Artificial

Intelligence: Economic Impact, Opportunities,

Challenges, Implications for Policy,” European

Economy Discussion Papers 210, European Union.

Takefuji, Y. (2024). “Unveiling inequality: A deep dive into

racial and gender disparities in US court case closures,”

Cities (154), p. 105398 (doi: 10.1016/j.citie

s.2024.105398).

vom Brocke, J., Simons, A., Niehaves, B., Reimer, K.,

Plattfaut, R., and Cleven, A. (2009). “Reconstructing

the Giant: On the Importance of Rigour in

Documenting the Literature Search Process,” in

Proceedings of the ECIS 2009, Verona, Italy.

08.06.2009 - 10.06.2009.

vom Brocke, J., Simons, A., Riemer, K., Niehaves, B.,

Plattfaut, R., and Cleven, A. (2015). “Standing on the

Shoulders of Giants: Challenges and Recommendations

of Literature Search in Information Systems Research,”

Communications of the Association for Information

Systems (37) (doi: 10.17705/1CAIS.03709).

Zoeten, M. C. de, Ernst, C.-P. H., and Rothlauf, F. (2024).

“The Effect of Explainable AI on AI-Trust and Its

Antecedents over the Course of an Interaction,” in ECIS

2024 Proceedings, Paphos, Cyprus. 13.06.2024 -

19.06.2024.

DATA 2025 - 14th International Conference on Data Science, Technology and Applications

660