Toolchain for Dataset Description with Contextual Recommendation

from Machine-Actionable Data Management Plans

Mohamed-Anis Koubaa

1 a

, Nan Liu

1 b

, Fabia Martens

1 c

, Andreas Schmidt

1,2 d

Karl-Uwe Stucky

1 e

and Wolfgang S

uß

1 f

Institute for Automation and Applied Informatics, Karlsruhe Institute of Technology, Germany

University of Applied Sciences, Karlsruhe, Germany

Keywords:

Research Data Management, Metadata, Automated Metadata Generation, Metadata Suggestions, Data

Description, FAIR Principles, RDMO, Data Management Plans, Ontologies, Semantic Web, Scientiﬁc Data,

Metadata Quality, Metadata Consistency, Digital Repositories, Information Systems, Knowledge Graphs.

Abstract:

By harnessing the power of interconnected research project information, this position paper introduces a novel

system designed to automate and enhance the metadata description process for research data. The system

effectively leverages existing structured data from RDMO (Research Data Management Organiser), drawing

insights from research projects, measurement equipment, sensors, and simulations to provide context-aware

suggestions for metadata ﬁelds. We argue that this system signiﬁcantly reduces the manual burden on re-

searchers, improves the quality and consistency of metadata, and ultimately champions the FAIR principles

(Findable, Accessible, Interoperable, Reusable) for all research data.

1 INTRODUCTION

The energy sector, with its foundation in long-

standing infrastructure and siloed data systems, faces

a monumental task in integrating disparate datasets

that are crucial for innovation and resilience. As

we shift towards decentralised energy production and

smart consumption, the challenge of managing re-

search data across these heterogeneous environments

becomes acute. High-quality, semantically rich meta-

data is not merely a best practice; it is the fundamen-

tal enabler of interoperability, bridging the gaps be-

tween legacy systems, new technologies, and multi-

stakeholder collaboration to drive the energy transi-

tion forward.

The process of creating metadata is largely man-

ual, time-consuming, and, unfortunately, prone to er-

rors. Researchers often spend considerable effort on

this administrative task, diverting valuable time and

resources away from their primary research activities.

https://orcid.org/0000-0001-8552-2008

https://orcid.org/0009-0005-8768-7072

https://orcid.org/0009-0007-2890-430X

https://orcid.org/0000-0002-9911-5881

https://orcid.org/0000-0002-0065-0762

https://orcid.org/0000-0003-2785-7736

This manual effort frequently leads to inconsistency

and signiﬁcant variability in metadata quality across

different datasets, projects, and even within the same

research group. Consequently, a critical problem

arises: a widespread lack of semantic interoperabil-

ity (Borgman, 2016). Without a common, machine-

readable understanding of the terms and concepts

used to describe data, it becomes exceedingly dif-

ﬁcult to integrate, compare, or reuse datasets effec-

tively, even when they originate from related studies

(Sasse et al., 2022). Ultimately, the burden on re-

searchers associated with this cumbersome, manual

metadata creation process not only impacts the qual-

ity of the metadata itself but also detracts signiﬁcantly

from core scientiﬁc discovery.

Many research institutions have already made sig-

niﬁcant investments in tools like Research Data Man-

agement Organiser (RDMO) (Anders et al., 2024)

to facilitate good research practices, particularly in

the creation and management of Data Management

Plans (DMPs). RDMO excels at guiding researchers

through the planning phase, prompting them to artic-

ulate crucial aspects such as data types, storage so-

lutions, access policies, and long-term preservation

strategies well in advance of data collection (Anders

et al., 2024). However, a persistent challenge re-

mains: the valuable, structured information metic-

506

Koubaa, M.-A., Liu, N., Martens, F., Schmidt, A., Stucky, K.-U. and Süß, W.

Toolchain for Dataset Description with Contextual Recommendation from Machine-Actionable Data Management Plans.

DOI: 10.5220/0013786300004000

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

In Proceedings of the 17th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2025) - Volume 2: KEOD and KMIS, pages

506-513

ISBN: 978-989-758-769-6; ISSN: 2184-3228

ulously documented within these DMPs often be-

comes siloed from the actual, hands-on data descrip-

tion phase. While a DMP outlines how data should

be managed, the practical application of this planning

– the creation of comprehensive metadata for the data

itself – frequently reverts to a separate, often man-

ual, process. This disconnection means that the rich

contextual information captured during the planning

stage is not automatically leveraged when the research

data is ultimately prepared for sharing, archiving, or

publication, leading to redundant efforts and potential

inconsistencies.

Our proposed system seamlessly integrates the

structured project and resource information that al-

ready resides within RDMO and offers a robust and

transformative solution. It tackles these pervasive

metadata challenges by utilising advanced seman-

tic reasoning capabilities. This innovative approach

promises to foster a signiﬁcantly more efﬁcient and

demonstrably higher-quality research data description

process, ultimately accelerating scientiﬁc discovery

and enhancing data reusability.

2 THE CURRENT LANDSCAPE

AND GAPS

While the idea of enriching datasets with descriptive

metadata is not new, many current metadata tools, es-

pecially those widely used in research, have signiﬁ-

cant drawbacks. These limitations contribute to the

persistent challenges in managing research data.

2.1 Limitations of Current Metadata

Tools

To guide data providers in linking metadata to their

actual data, generic forms and templates are a stan-

dard tool. However, despite offering a basic struc-

ture, these tools frequently lack the necessary speci-

ﬁcity and ﬂexibility to capture the rich, nuanced con-

text inherent in diverse scientiﬁc data. This forces

researchers to manually enter information, often via

free-text descriptions, which ultimately results in in-

consistencies and ambiguities in the metadata.

Furthermore, these tools frequently suffer from a

lack of contextual awareness. They operate as a stand-

alone data entry systems, which is disconnected from

the broader research ecosystem. They do not inher-

ently ”know” about the project’s funding details, the

precise speciﬁcations of the sensor that generated the

data, or the parameters of the simulation software

used, even if such information is meticulously doc-

umented elsewhere. This forces researchers to heav-

ily rely on manual input, not just for the data itself,

but for re-entering or re-interpreting information that

already exists in other structured formats. The ab-

sence of automated linkages and intelligent sugges-

tions means the burden of connecting disparate pieces

of information falls squarely on the individual re-

searcher, perpetuating the inefﬁciencies and quality

issues previously identiﬁed.

Provenance data captured during workﬂow execu-

tion offers a rich source for ”Provenance” and ”Con-

text” metadata in dataset publication (Alper et al.,

2013). However, directly using this raw data can in-

advertently expose internal operational details, which

are irrelevant and potentially confusing to data con-

sumers. Modern data orchestrators and sequencers,

such as Dagster, solve this by providing a more

declarative and curated approach, offering a higher-

level view of data lineage and semantic context tai-

lored for external consumption.

2.2 Efforts Towards

Machine-Actionable DMPs

The Research Data Alliance (RDA) is a key player in

making efforts to generate a common understanding

for machine-actionable DMPs, which are very helpful

for the automation of annotation processes. Specif-

ically, the RDA Common Application Programming

Interface (API) for machine-actionable Data Man-

agement Plans Working Group is actively developing

this API speciﬁcation.

This project is built on the foundation of these

advancements and aligns with established principles

for machine-actionable DMPs, outlined in (Miksa

et al., 2019). Speciﬁcally, the following principles are

implicitly integrated as foundational system require-

ments for the present work:

Principle 1. Integrate DMPs with the workﬂows of

all stakeholders in the research data ecosystem

Principle 3. Make policies (also) for machines, not

just for people

Principle 5. Use Persistent Identiﬁers (PIDs) and

controlled vocabularies

Principle 6. Follow a common data model for

maDMPs

Principle 8. Support data management evaluation

and monitoring

Principle 9. Make DMPs updatable, living, ver-

sioned documents

Principle 10. Make DMPs publicly available

Toolchain for Dataset Description with Contextual Recommendation from Machine-Actionable Data Management Plans

507

2.3 Energy-Domain Speciﬁc Tools

In the speciﬁc context of the energy sector, no-

table efforts are underway to streamline meta-

data creation. The Open Energy Platform (OEP)

(openenergyplatform.org), for example, provides a

web-based metadata wizard designed to guide users

through the process of describing their energy-related

datasets. This wizard is an important step forward

as it simpliﬁes metadata entry, enforcing a speciﬁc

metadata standard (OEMetadata, which is built on

Frictionless Data Package speciﬁcations and aligns

with Findable, Accessible, Interoperable & Reusable

(FAIR) principles) and assisting users in providing

crucial context for their data. It also integrates

with the Open Energy Ontology (OEO) introduced in

(Booshehri et al., 2021), enabling semantic annota-

tion of data and enhancing ﬁndability.

However, while the OEP’s wizard brings signiﬁ-

cant improvements upon generic forms by providing

a structured, domain-speciﬁc approach, it still primar-

ily relies on the researcher’s direct manual input into

many of its ﬁelds. It guides what information to pro-

vide, but is not able to suggest values for those ﬁelds.

A researcher might still need to manually re-enter

project titles, funding details, or sensor calibration pa-

rameters, even if these were already documented in an

RDMO Data Management Plan or an internal lab in-

ventory system. This shows a potential area where

the already upstream captured information is not au-

tomatically leveraged, necessitating redundant efforts

and potentially leading to inconsistencies if the man-

ual entry differs from the details of DMP.

Apart from the Open Energy Platform’s metadata

wizard, other initiatives directly address energy data

sharing. EnergySHR (https://energyshr.nl) as an ex-

ample serves as a specialised platform explicitly de-

signed for energy dataset sharing and communica-

tions, particularly focusing on researchers working on

AI and data-driven solutions for the energy transition.

While EnergySHR and similar platforms are cru-

cial for breaking down data silos and fostering col-

laborative research by providing the infrastructure for

data exchange, their primary focus often lies on the

sharing mechanism and access control rather than the

initial automated generation or intelligent suggestion

of comprehensive metadata.

2.4 RDMO’s Role and Limitations in

Isolation

While RDMO has established itself as an indispens-

able tool for many institutions by enabling the struc-

tured creation and systematic collection of informa-

tion for DMPs (Mozgova et al., 2022), a signiﬁ-

cant gap persists between this detailed planning phase

(DMPs creation) and the subsequent practical meta-

data application during the actual data description

process (Windeck et al., 2024). The structured de-

tails about a project, the speciﬁc measurement equip-

ment used, the characteristics of sensors, or the pa-

rameters of simulations — all recorded in RDMO —

are frequently not used for the subsequent task of cre-

ating descriptive metadata for the research output it-

self. When preparing data for publication, sharing, or

long-term archiving, researchers often resort to man-

ually re-entering or re-interpreting the same informa-

tion into separate metadata forms or schemas. This

disconnection causes redundant work, risks inconsis-

tencies, and isolates valuable context within the DMP,

as planning details are not automatically used to en-

hance the ﬁnal metadata, reducing the efﬁciency and

quality beneﬁts intended by RDMO.

2.5 Addressing the Gaps

Simple, static metadata forms, while offering a basic

framework, prove insufﬁcient for truly effective re-

search data description in today’s complex scientiﬁc

landscape. For instance, on the OEP, the vast ma-

jority of tables lack semantic annotations at both the

subject and ﬁelds’ descriptions, with only 23% pos-

sessing such terms. Their fundamental limitation lies

in their passive nature.

To overcome these identiﬁed gaps, intelligent,

context-aware suggestions are not just beneﬁcial, but

necessary. This shifts the researcher’s role from labo-

rious data entry to efﬁcient validation and reﬁnement,

dramatically increasing the speed, accuracy, and rich-

ness of metadata, thereby making data truly FAIR

and unlocking its full potential for future scientiﬁc in-

quiry. See Table 1.

3 PROPOSED SYSTEM:

AUTOMATED METADATA

SUGGESTION

At its core, the proposed system introduces a trans-

formative approach to metadata creation: it leverages

the already captured rich and structured data within

RDMO – speciﬁcally regarding research projects,

measurement equipment, sensors, and simulations –

and combines it with robust semantic knowledge to

automatically generate intelligent suggestions for var-

ious metadata ﬁelds. This fundamental concept aims

to move beyond manual input, providing researchers

KMIS 2025 - 17th International Conference on Knowledge Management and Information Systems

508

Table 1: Comparison of current tools and the proposed system.

Criteria Current Tools Proposed System

Functionality

and Automation

Manual metadata entry, often in generic

templates and in free-text; prone to (hu-

man) error and inconsistencies

Collects information from external sources

and provides suggestions. Uses existing

knowledge from various sources, domain-

speciﬁc semantics, consistency with pre-

existing plans and scientiﬁc standards.

Semi-automatic preﬁlled ﬁelds based on

context, the speciﬁc project, data type, in-

volved equipment, etc.

Time time-consuming manual labour, no pre-

ﬁlled ﬁelds

time-efﬁcient automatisation, only de-

mands review (validation and reﬁnement)

of the preﬁlled ﬁelds

Interoperability Low, alone-standing, often platform-

speciﬁc or unique formats

Supports API, RO-Crate, JSON-LD

Integration Low to none, often not able to read JSON-

LD, no contextual awareness

High, access to many different sources,

creates machine-readable output, processes

JSON-LD

Standards Com-

pliance

Differs FAIR, JSON-LD

with contextually relevant and accurate metadata rec-

ommendations from the outset.

3.1 Core Functionalities

At the heart of this system lie several core functional-

ities designed to ensure its effectiveness and seamless

operation:

Access Available Sources of Information (CF1)

The system’s foundational capability is its ability to

access and retrieve data from all relevant information

sources. This includes structured data from RDMO,

as well as potentially other institutional databases, ex-

ternal registries, and semantic resources.

Present Information in Suitable Form to

System-Subcomponents (CF2)

Once accessed, information is processed and trans-

formed. This core function ensures that the retrieved

data is presented in a standardised and suitable for-

mat, enabling efﬁcient consumption and utilisation by

all other internal system sub-components, such as the

Knowledge Base and the Suggestion Engine.

Present an Easy Way to Edit Metadata for the

Editor, Publisher and Curator (CF3)

Recognising that automated suggestions are a pow-

erful starting point but not without their risks for er-

rors or misinterpretation, a fundamental functional-

ity is providing a user-friendly interface for curators.

This ensures an easy and intuitive way for humans

to review, modify, and reﬁne the suggested metadata,

maintaining oversight and guaranteeing accuracy.

Use Programming Interfaces of Target Platforms

to Publish and Search Metadata (CF4)

For the generated and curated metadata to be truly

impactful, the system must actively interact with the

broader research data ecosystem. This functionality

uses programming interfaces (APIs) of target plat-

forms, such as institutional repositories or data por-

tals, to seamlessly publish the rich metadata and en-

able efﬁcient searching, thereby maximising the ﬁnd-

ability and reusability of the research data.

3.2 Key Components and Core

Functions Coverage

This system can be broken down into several inter-

connected components:

User Interface (UI) for Data Description is the

primary interface where researchers can input and

review metadata. It has to be intuitive, guided,

and interactive. The core function CF3 requires

that the suggestions are made available in the UI,

to guide the user in selecting a meaningful value

for the attribute or property being edited.

RDMO Integration Module connects to the

RDMO instance to retrieve project information

(e.g. project title, abstract, principal investigator,

funding, associated DMPs), and details about

Toolchain for Dataset Description with Contextual Recommendation from Machine-Actionable Data Management Plans

509

measurement equipment, sensors, or simulations

that are already documented in RDMO. More

generally, modules are mainly integrated via

APIs-based connections. Data transferred via

API receives further semantic transformation

when the source and sink schemas are known and

independent yet interoperable via mapping.

Knowledge Base with Semantic Layer is the brain

of the suggestion system. It contains the rules,

relationships, and semantic understanding neces-

sary to generate intelligent suggestions.

Suggestion Engine (SE) takes input from the user

(partially ﬁlled metadata and context from

RDMO) and queries the Knowledge Base to gen-

erate suggestions.

External Data Source Integrator retrieves infor-

mation from other relevant systems beyond

RDMO that might hold valuable metadata (e.g.

lab inventory systems, institutional repositories,

external registries of scientiﬁc instruments, etc.).

Metadata Repository is used to store the fully en-

riched metadata (after the user input and sugges-

tions are ﬁnalised).

3.3 Workﬂow

The user’s journey begins within the RDMO’s intu-

itive and highly ﬂexible user interface, designed to

allow for extensive customisation through Question-

naire Deﬁnitions. These deﬁnitions are managed by a

superuser (see 1 in Figure 1) and are crafted to tailor

the questions presented to researchers (see 2 ). Cru-

cially, these questionnaires have a deﬁned data model

that can be directly mapped and utilised in subsequent

stages of the research data management workﬂow, en-

suring seamless data ﬂow and consistency.

3.3.1 Initiating Data Description

A researcher, having completed their data collection

or simulation, navigates to the system to begin de-

scribing their new dataset. They select the relevant

RDMO project from a pre-populated list (populated

by the RDMO Integration Module). This initial se-

lection immediately provides the system with crucial

context.

3.3.2 Initial Contextual Pre-Population

(RDMO-driven)

Upon selecting the project, the system instantly pre-

populates (in 3 ) several core metadata ﬁelds based

on the information already available in the associated

RDMO DMP. This includes:

1. Project Title, Abstract, and Principal Investigator,

2. Funding information (grant numbers, funding

agencies) and

3. High-level data types (e.g. ”experimental data” or

”simulation output”).

Furthermore, the system broadcasts important in-

formation (in 4 ) to further software components

(e.g. Electronic Lab Notebookss (ELNs)) and facil-

itates, thereby, a potentially data description step (in

5 ). The concrete research operation produces the re-

sulting dataset ( 6 ) and makes it available for further

processing by the toolchain.

3.3.3 Receiving Contextual Suggestions (RDMO

and Semantic Knowledge)

During the process, the Suggestion Engine springs

into action whenever ﬁelds are not directly pre-

populated and infers both the RDMO data and its own

integrated Knowledge Base:

• If the RDMO project details indicate the use of

a speciﬁc measurement equipment, the system

proactively suggests relevant instrument types,

manufacturer details, and common measurement

units or parameters associated with that equip-

ment.

• Similarly, if RDMO outlines the use of particu-

lar sensors, the system suggests appropriate units

(Celsius, Kelvin, etc.), temporal resolution, or rel-

evant calibration procedures.

• For simulation data, suggestions are based on the

software documented in RDMO and include soft-

ware names, versions, relevant physical constants

or models used in the simulation.

• For ﬁelds like ”Keywords” or ”Scientiﬁc Disci-

pline,” the system suggests terms extracted from

the project abstract in RDMO or broader ontolog-

ical terms related to the project’s primary domain.

3.3.4 Interacting with Suggestions

Suggestions are presented in a clear, non-intrusive

manner as highlighted text, auto-complete drop-

downs, or clickable buttons next to the relevant ﬁeld.

The researcher can then:

1. Accept a suggestion with a single click, instantly

populating the ﬁeld.

2. Modify a suggestion if it is close but not perfectly

accurate.

3. Ignore/Reject a suggestion and, if preferred, man-

ually enter their own value.

KMIS 2025 - 17th International Conference on Knowledge Management and Information Systems

510

3.3.5 Reﬁnement and Validation

As the researcher (in 7 ) ﬁlls in more ﬁelds, the sys-

tem can dynamically reﬁne its suggestions, leveraging

the newly entered information for even greater accu-

racy.

Before the ﬁnal submission (

8 and 9 ), the sys-

tem performs validation checks, ensuring all manda-

tory ﬁelds are complete and values adhere to estab-

lished standards (e.g. valid date formats, correct unit

types).

The workﬂow seamlessly integrates planning in-

formation with intelligent, context-aware suggestions.

Several semantic transformations are inherently part

of the different system interfaces and are a core func-

tion of the proposed system.

4 ADVANTAGES AND IMPACT

The adoption of the proposed system yields a mul-

titude of signiﬁcant advantages, primarily aimed at

reforming research data management practices. The

foremost among these are substantial efﬁciency gains,

which directly translate into a marked reduction in the

time researchers must dedicate to manual metadata

entry. By automating large portions of this process,

the system effectively frees valuable researcher time,

allowing them to redirect their focus back to core sci-

entiﬁc inquiry and innovation.

Beyond mere efﬁciency, the system enables a dra-

matic improvement in metadata quality and consis-

tency. The intelligent suggestion mechanism, draw-

ing from structured RDMO data and semantic knowl-

edge, inherently leads to the minimisation of errors

and omissions that frequently plague manual input.

Moreover, it actively promotes adherence to estab-

lished standards and controlled vocabularies, moving

away from free-text ambiguity towards standardised,

machine-readable descriptions. This, in turn, funda-

mentally enhances semantic interoperability, ensuring

that data from different sources can be seamlessly in-

tegrated and understood across various platforms and

disciplines.

The proposed system directly contributes to the

promotion of the FAIR Principles for research data.

Its automated (like encouraged in Rec. 8 from (Eu-

ropean Commission. Directorate General for Re-

search and Innovation., 2018), p. 45) and standard-

ised metadata generation capabilities make data inher-

ently more Findable by search engines and data cat-

alogues, more Accessible through consistent descrip-

tions, more Interoperable due to semantic alignment,

and ultimately more Reusable by a broader scientiﬁc

community.

Furthermore, a key strength of this approach lies

in its ability to leverage existing infrastructure. In-

stead of requiring a complete overhaul of institutional

data management strategies, the system intelligently

builds upon and extends existing investments in tools

like RDMO. This not only optimises resource utilisa-

tion but also signiﬁcantly increases the overall value

proposition of these pre-existing data management

planning efforts.

Finally, the system is designed with scalability

and adaptability in mind. Its modular architecture

and reliance on conﬁgurable semantic rules mean it

possesses considerable potential for straightforward

adaptation to the unique metadata requirements of

various scientiﬁc disciplines, as well as for accommo-

dating evolving metadata standards and best practices

over time. This ensures its long-term relevance and

utility in a rapidly changing data landscape.

5 CONCLUSION AND FUTURE

OUTLOOK

At its core, the proposed system represents a crucial

step forward in research data management. By intel-

ligently leveraging existing information from RDMO

and integrating it with robust semantic knowledge,

it directly reduces the inefﬁciencies, inconsistencies,

and manual burdens widespread in current metadata

creation. This system not only streamlines the pro-

cess but also elevates the quality and interoperability

of research data, ultimately fostering a more efﬁcient

and effective environment for scientiﬁc discovery and

data reusability.

Building on the robust foundation of the proposed

system, several exciting paths for future research and

development open, promising to further enhance its

intelligence, reach, and utility in research data man-

agement. The following sections present an overview.

Incorporating ML for More Advanced

Suggestions

While the initial system relies on rule-based semantic

reasoning, integrating machine learning (ML) models

can unlock a new level of intelligence. This includes:

Natural Language Processing (NLP). Analysing

free-text descriptions within RDMO (e.g. project

abstracts, methodology sections) or existing meta-

data to extract entities, topics, and relationships

that can improve suggestions.

Recommendation Proﬁles. Learning from patterns

in previously described datasets and user interac-

Toolchain for Dataset Description with Contextual Recommendation from Machine-Actionable Data Management Plans

511

Figure 1: System Key Components with References to Workﬂow-Steps.

tions (e.g. accepted/rejected suggestions) to pro-

vide highly personalised and accurate recommen-

dations for metadata values. This can address

the ”cold start” problem for new types of data or

equipment.

Anomaly Detection. Using Machine Learning (ML)

to identify potential inconsistencies, gaps, or un-

usual values in metadata, prompting users to re-

view and correct. First approaches are shown in:

(Wu et al., 2023).

Active Learning. Implementing active learning

strategies where the system intelligently queries

the user for input on uncertain suggestions,

for continuously reﬁning and improving its

underlying models. This creates a powerful

human-in-the-loop learning mechanism.

Deeper Integration with ELNs and Laboratory

Information Management Systems (LIMS)

ELNs and LIMS are where much of the raw exper-

imental and observational data, along with detailed

methodologies, are ﬁrst recorded. Future work could

focus on:

Automated Metadata Extraction from ELN

Entries. Developing parsers and NLP models

to automatically extract relevant metadata (e.g.

instrument settings, reagent batch numbers, en-

vironmental conditions, experimental protocols,

etc.) from structured and unstructured ELN

entries.

Real-Time Metadata Flow. Establishing seamless,

automated pipelines for metadata to ﬂow directly

from ELNs and LIMS into the metadata sugges-

tion system, drastically reducing manual entry

post-experiment.

Linking Data Provenance. Enriching metadata

with a detailed provenance trail directly from

ELNs or data orchestrators, linking speciﬁc data

points back to the exact experimental steps and

equipment used.

Integration with Other Research Information

Systems

Integrating with other existing research information

systems offers signiﬁcant opportunities to automat-

ically enrich metadata with contextual details that

are often difﬁcult to track manually. This includes

leveraging information from Current Research Infor-

mation Systems (CRIS) or ORCID proﬁles to au-

tomatically suggest accurate author details, afﬁlia-

tions, and persistent researcher IDs, ensuring proper

attribution and discoverability. The system could

also connect to external or institutional instrument

registries, pulling detailed speciﬁcations, calibration

data, and even maintenance logs for speciﬁc equip-

ment or sensors, thereby greatly enriching the meta-

data associated with experimental data. Furthermore,

for data derived from simulations or analyses, inte-

grating with software repositories like GitHub or in-

stitutional code archives allows the system to auto-

matically suggest crucial details such as software ver-

sions, dependencies, and licensing information, en-

suring transparency and reproducibility.

Advanced Semantic Reasoning and Knowledge

Graph Expansion

Future developments aim to extend beyond basic data

mappings by integrating more advanced inferencing

and enrichment mechanisms. An example would be a

system that is able to recognise particular sensor types

KMIS 2025 - 17th International Conference on Knowledge Management and Information Systems

512

and then automatically infers the common measure-

ments or the associated physical quantities. Even if

not explicitly stated, the system can capture the infor-

mation and reduce manual labour.

User Experience (UX) and Interactive Feedback

Mechanisms

Beyond core functionality, reﬁning the user experi-

ence is paramount for adoption and continuous im-

provement. A key area is Explainable AI for Sugges-

tions, where the system provides clear, concise ex-

planations for its recommendations. Transparent ex-

planations like ”This unit was suggested because it

is commonly associated with this sensor type, which

is listed in your RDMO DMP for this project.” are

important to build user trust and aid understanding.

Furthermore, exploring gamiﬁcation and incentives

could encourage researchers to adhere to metadata

best practices and provide valuable feedback, trans-

forming a potentially tedious task into an engag-

ing one. Finally, developing features for collabora-

tive metadata curation would empower multiple re-

searchers or data stewards to jointly review and reﬁne

suggested metadata, especially for complex or inter-

disciplinary datasets, fostering a shared sense of own-

ership and accuracy.

ACKNOWLEDGEMENTS

This publication was supported by the Helmholtz

Metadata Collaboration (HMC), an incubator plat-

form of the Helmholtz Association within the frame-

work of the Information and Data Science strategic

initiative.

The authors would like to thank the German Fed-

eral Government, the German State Governments,

and the Joint Science Conference (GWK) for their

funding and support as part of the NFDI4Energy

consortium. The work was funded by the Ger-

man Research Foundation (DFG) – 501865131 within

the German National Research Data Infrastructure

(NFDI, www.nfdi.de).

REFERENCES

Alper, P., Belhajjame, K., Goble, C. A., and Karagoz, P.

(2013). Enhancing and abstracting scientiﬁc workﬂow

provenance for data publishing. In Proceedings of the

Joint EDBT/ICDT 2013 Workshops, pages 313–318,

Genoa Italy. ACM.

Anders, I., Enke, H., Hausen, D. A., Henzen, C., Jagusch,

G., Lanza, G., Michaelis, O., Peters-von Gehlen, K.,

Rathmann, T., Rohrwild, J., Sch

onau, S., Wedlich-

Zachodin, K. V., and Windeck, J. (2024). The Re-

search Data Management Organiser (RDMO) – a

Strong Community Behind an Established Software

for DMPs and Much More. Data Science Journal,

23:28.

Booshehri, M., Emele, L., Fl

ugel, S., F

orster, H., Frey,

J., Frey, U., Glauer, M., Hastings, J., Hofmann, C.,

Hoyer-Klick, C., H

ulk, L., Kleinau, A., Knosala, K.,

Kotzur, L., Kuckertz, P., Mossakowski, T., Muschner,

C., Neuhaus, F., Pehl, M., Robinius, M., Sehn, V., and

Stappel, M. (2021). Introducing the Open Energy On-

tology: Enhancing data interpretation and interfacing

in energy systems analysis. Energy and AI, 5:100074.

Borgman, C. L. (2016). Big data, little data, no data:

scholarship in the networked world. The MIT Press,

Cambridge, Massachusetts London, England, ﬁrst mit

press paperback edition edition.

European Commission. Directorate General for Research

and Innovation. (2018). Turning FAIR into reality: ﬁ-

nal report and action plan from the European Com-

mission expert group on FAIR data. Publications Of-

ﬁce, LU.

Miksa, T., Simms, S., Mietchen, D., and Jones, S.

(2019). Ten principles for machine-actionable data

management plans. PLOS Computational Biology,

15(3):e1006750.

Mozgova, I., Jagusch, G., Freund, J., Kraft, A., Gl

uck,

T., Herrmann, K., Kn

ochelmann, M., and Lachmayer,

R. (2022). Product Life Cycle Oriented Data Man-

agement Planning with RDMO at the Example of

Research Field Data. In E-Science-Tage 2021. hei-

BOOKS.

Sasse, J., Darms, J., and Fluck, J. (2022). Seman-

tic Metadata Annotation Services in the Biomedical

Domain—A Literature Review. Applied Sciences,

12(2):796.

Windeck, J., Schr

oder, M., Frenzel, J., Kusmierz,

S., Fuchs, H., and Klar, J. (2024). Bausteine

Forschungsdatenmanagement : 2024, 2Connecting

RDM Services. Bausteine Forschungsdatenman-

agement, (2). Medium: application/pdf Publisher:

Philipps-Universit

at Marburg.

Wu, M., Brandhorst, H., Marinescu, M.-C., Lopez, J. M.,

Hlava, M., and Busch, J. (2023). Automated metadata

annotation: What is and is not possible with machine

learning. Data Intelligence, 5(1):122–138.

Toolchain for Dataset Description with Contextual Recommendation from Machine-Actionable Data Management Plans

513