PaperlayerView exact revision on arXiv

Source-linked AI summary

From Queries to Narratives: Cultural Heritage Data Stories for Knowledge Graph Exploration and Quality Assessment

Tabea Tietz, Torsten Schrade, Etienne Posthumus, Linnaea Söhn, Jonatan Jalle Steller, Jörg Waitelonis, Harald Sack

arXiv:2609.11403v1cs.AIcs.DL

TL;DR

Users can struggle to formulate meaningful queries against specialized, heterogeneous, and growing cultural-heritage knowledge graphs. The paper presents data stories and the LODEON authoring environment as ways to guide querying and exploration; early experience is positive, while the authors stress that AI limits need to be transparent.

  • Problem

    Users may lack the graph-model knowledge and starting points needed to discover what a specialized, heterogeneous, growing cultural-heritage knowledge graph contains or query it meaningfully.

    Source: §1 · p. 1

  • Method

    The paper presents data stories combining explanatory narrative with executable queries and results, and introduces LODEON with Sparnatural and an AI assistant for authoring against a knowledge graph.

    Source: §1 · p. 2, §4.3 · p. 8

  • Results

    Early experience suggests LODEON is useful and widely received positively, with its AI assistant lowering the query-construction barrier and helping participants understand an unfamiliar graph.

    Source: §6 · p. 11

  • Takeaways & Limitations

    Authoring data stories can surface data-quality issues that remain hidden in aggregate statistics.

    Source: §1 · p. 2

    AI assistance is not sufficient without clearer transparency about potentially wrong results and the limits imposed by infrastructure as well as data.

    Source: §4.3.2 · p. 10

Abstract

from arXiv · show

Cultural-heritage KGs such as the NFDI4Culture-KG contain millions of triples about artworks, music, inscriptions, historical events, and the people and places connected to them. For many users, however, discovering this knowledge can be difficult. While SPARQL can be learned, writing meaningful queries first requires an in-depth understanding of the graph's data model, an investment many domain researchers and practitioners are unwilling to make. Even with existing user interfaces, a starting point and some guidance are usually needed, because the data contained in the graph is highly specialized, heterogeneous, and constantly growing, making it challenging to know what it contains or which questions it can answer. In this paper, we present data stories as a way not only to lower this barrier, but also to turn exploration into data-quality assessment, and thus combine accessible querying with the discovery of issues that remain hidden in aggregate statistics. In this contribution, a data story is understood as a narrative document that integrates explanatory text and images with executable SPARQL queries and their visualized results. It is described how they are authored against the graph and how they serve several purposes: guiding users through an unfamiliar graph, creating reproducible narratives, and surfacing data-quality issues previously hidden in aggregate statistics. The authoring platform LODEON including its Sparnatural and AI-supported authoring assistants is introduced as a proof-of-concept. Within the authoring environment, every claim made about the data can be backed by an explicit query, making these narratives transparent and reproducible. This paper also reflects on lessons learned from hands-on seminars and workshops. Early experience suggests that such data stories make cultural-heritage knowledge graphs more accessible for both exploration and quality assessment.

1. Introduction

The NFDI4Culture-KG offers a large, heterogeneous cultural-heritage resource, but its specialized data model and evolving contents make it difficult for many users to know what it contains or how to query it. The paper presents data stories as guided, reproducible exploration that can also surface data-quality issues, and introduces LODEON as an AI-assisted authoring proof-of-concept.

1. Introduction

  • The graph’s scale, heterogeneous sources, and layered ontologies make meaningful queries difficult for users who lack time to learn its data model.The paper describes approximately 150 million triples and identifies the Culture Ontology, NFDIcore, and BFO as its ontology layers.Source: §1 · pp. 1–2
  • A data story combines explanatory text and images with executable SPARQL queries and visualized results to guide exploration and expose quality issues.The paper describes stories as onboarding tools, starting points for users’ own questions, and a way to surface issues hidden in aggregate statistics.Source: §1 · p. 2
  • The paper presents LODEON with Sparnatural and an AI authoring assistant as a proof-of-concept, motivated by friction observed in teaching and workshops.The earlier NFDI4Culture environment was realized with SHMARQL and used in teaching and international workshops.Source: §1 · p. 2
  • The use cases center on the NFDI4Culture-KG and, in part, the German Memory Atlas, although LODEON is described as applicable to any knowledge graph.The paper frames access, query formulation, and reproducibility as challenges beyond cultural heritage.Source: §1 · p. 2

2. Related Work

Prior work offers narrative presentations of data or tools for exploring and querying knowledge graphs, while this paper combines narrative guidance with query access over a live cultural-heritage graph.

2.1. Data Stories and Narrative Visualization

  • Existing narrative research interfaces make data approachable, but stories are typically based on fixed exports rather than inspectable, rerunnable queries against live data.The comparison includes the CLARIAH Data Stories Editor and Carnegie Hall Data Lab experiments.Source: §2.1 · pp. 2–3

2. Related Work

  • The contribution combines SHMARQL, Sparnatural, and an LLM-based assistant within a storytelling format over a live cultural-heritage knowledge graph.The format lets readers inspect and rerun the query behind a claim, with results reflecting the current data rather than a fixed snapshot.Source: §2.2 · p. 3, §2.1 · p. 3

2.2. Querying and Exploring Knowledge Graphs

  • Knowledge-graph exploration tools support navigation, SPARQL querying, visual query building, or natural-language-to-SPARQL translation, but do not by themselves provide narrative guidance.Examples include Sparnatural and LLM-based translation systems.Source: §2.2 · p. 3

3. Bridging the Distance with Data Stories in NFDI4Culture

NFDI4Culture-KG integrates decentralized cultural-heritage resources, but users need concrete examples of what the graph contains and which research questions it can answer. Data stories pair questions with working queries to make graph conventions visible and support knowledge exchange.

3. Bridging the Distance with Data Stories in NFDI4Culture

  • NFDI4Culture-KG is a semantic index that provides a single access point to cultural-heritage research data while the contributing institutions retain the data.It is modeled with the Culture Ontology, which extends NFDIcore and is aligned with BFO.Source: §3 · p. 3
  • The graph contains more than 3.7 million resources from 18 provider feeds, with potential to grow across NFDI4Culture’s 88 listed data portals.Provider data is processed through the Culture Kitchen ETL pipeline.Source: §3 · p. 3
  • Users need examples of graph contents and research questions, plus working queries that they can read, adapt, and cite as part of their research process.Ontology documentation, a public SPARQL endpoint, and search interfaces alone have not resolved these needs.Source: §3 · pp. 3–4
  • Data stories pair specific questions with executable queries to make data conventions visible and help close the distance between a documented schema and research questions.The paper presents this knowledge-exchange role as relevant to heterogeneous, continuously growing knowledge graphs beyond NFDI4Culture.Source: §3 · p. 4

4. Data Stories in NFDI4Culture from Consumption to Creation and AI Assisted Exploration

This section follows data stories from reading existing narratives to authoring them, then introduces LODEON and reports early teaching and workshop experience.

4.1. Consuming Data Stories as an Entry Point to Structured Data

  • Existing stories pair research questions with queries to reveal graph conventions, illustrate reusable query patterns, and show how heterogeneous feeds can be linked.“An Italian Data Journey” moves from a portal-and-record overview to analyses including opera-premiere mapping; another story demonstrates linking musical sources and Gregorovius letters.Source: §4.1 · pp. 4–5

4.2. Authoring Data Stories: A SHMARQL-based Environment

  • SHMARQL executes SPARQL within Markdown narratives and keeps visualizations tied to live, editable queries, making each story claim inspectable.Readers can modify queries to explore further or inspect data provenance.Source: §4.2 · p. 6

4.2.1. Lessons Learned on SHMARQL-supported Story Creation

  • KIT students learned graph modeling patterns through goal-directed story authoring, but writing SPARQL from scratch remained a hurdle for beginners.After initial onboarding, students successfully queried and visualized data; this friction motivated more active query-construction support.Source: §4.2.1 · p. 7

4.3. Towards AI Assisted Authoring with LODEON

  • LODEON combines KG interaction and note-taking with Sparnatural and AI-supported query assistance, while letting authors select and control story content.Its assistant can inspect the active graph profile, look up ontology terms, validate and run queries, and propose queries or visualizations; stories can also include specialized media nodes and metadata for export or publication.Source: §4.3 · pp. 7–9

4.3.1. Availability of LODEON

  • The public workbench does not yet offer LODEON’s AI assistant freely; users can run the open-source tool locally and connect their own models.The paper attributes the public-workbench restriction to associated costs.Source: §4.3.1 · p. 9

4.3.2. Lessons Learned on LODEON and AI Assisted Data Exploration

  • At a hands-on DH2026 workshop with roughly 60 participants, LODEON was widely judged useful for cultural heritage and digital humanities work, while AI-result ratings mostly clustered between 2 and 4 on a 1-to-10 scale.AI answers were particularly unreliable for aggregate questions such as counts, yet looked equally trustworthy whether correct or not.Source: §4.3.2 · pp. 9–10
  • Workshop experience suggests AI helps users approach an unfamiliar graph only partially and under conditions; explicit queries alone do not ensure transparent, trustworthy answers.The authors call for clearer indications of potentially wrong results and of infrastructure limits that affect answers.Source: §4.3.2 · pp. 9–10

5. Quality Assessment

The section presents story authoring as a complement to aggregate KG quality measures and reports issues found through hands-on exploration.

5. Quality Assessment

  • Aggregate quality measures work at scale but can miss localized feed problems; story authoring exposes such issues through retrieval and inspection of concrete records.The paper frames story authoring as a complement to, not a replacement for, automated quality assessment.Source: §5 · p. 10
  • KIT seminar students uncovered missing data points and controlled-vocabulary entries while authoring stories, and some missing fields were added in later ingestion rounds.The missing vocabulary entries were identified as necessary for standardized linking.Source: §5 · p. 10
  • Issue reporting currently uses GitHub issues or contact with NFDI4Culture, while a structured reporting mechanism, including for missing connections, is planned.The planned reports may also reveal requirements for the underlying ontology.Source: §5 · p. 10

6. Conclusion

The conclusion presents data stories as a way to make the NFDI4Culture-KG more accessible and use exploration for quality assessment. It describes LODEON as a proof-of-concept and reports encouraging early experience alongside planned evaluation and development.

6. Conclusion

  • LODEON extends SHMARQL with Sparnatural and an AI assistant as a proof-of-concept workbench.The conclusion frames the paper as following data stories from reading to authoring.Source: §6 · pp. 10–11
  • Early experience suggests data stories are useful, and LODEON's reception is widely positive.The AI assistant reportedly lowers the barrier to query construction and helps participants understand an unfamiliar graph.Source: §6 · p. 11
  • Next steps include quantitative evaluation of query accuracy and model performance and cost, followed by further development before official release.Future development will also consider use cases in Materials Science and Sports Science.Source: §6 · p. 11
  • The AI assistant can bridge users to the data in research only when its limitations are made transparent.Source: §6 · p. 11

Declaration on Generative AI

The authors report using Claude Opus 4.8 for language editing and state that they reviewed and edited the content afterward.

Declaration on Generative AI

A. LODEON Features

LODEON supports embedding remote content, illustrated by an interactive 3D model from the semantic kompakkt data portal.

A. LODEON Features

Loading 2609.11403v1…