Bigger Smarter Data

Extracting, Modeling and Linking Data for Literary History

Christof Schöch
(Trier University, Germany)

Korea University
Seoul, South Korea

23 May 2024

Introduction

Thanks

Seung-eun Lee of Korea University / Department of Korean Language and Literature / Humanities Utmost Sharing System, and Byungjun Kim, KAIST, on behalf of KADH (Korean Association for Digital Humanities).

The Ministry for Research, Education and Culture of Rhineland-Palatinate, Germany, for funding this research (Mining and Modeling Text, 2019-2023)

Thanks to all the project contributors: Maria Hinzmann, Matthias Bremm, Tinghui Duan, Anne Klee, Johanna Konstanciak, Julia Röttgermann and many others.

Overview

• 1 – Introduction
• 2 – Bigger Smarter Data and Linked Open Data
• 3 – Mining: Information Retrieval
• 4 – Modeling: Linked Open Data
• 5 – Results: Queryable Database
• 6 – Conclusion

• Hello and welcome everyone.
• It is great to be here!
  
• My topic today is a vision of Digital Humanities that tries to be both bigger and smarter, without choosing between the two.
• To do so, I will show how what “Linked Open Data” is.
• And how LOD can be one way of achieving “bigger smarter data”
• Specifically, I will use the way we have used LOD in a project on creating a new way of engaging with literary history
• I hope it will be interesting to you even, whether you are a literary historian, a scholar of history, or just curious about Digital Humanities.
• First, I will explain what I mean by “bigger smarter data” and by linked open data.
• Then I will show how we used text mining to collect information on French literary history.
• And how we used LOD to model this data and link it together, as well as link it to other resources.
• At the end, I will show what kinds of questions can be addressed using the resulting data.

Bigger Smarter Data:
Linked Open Data

Three Modes of Data (and Digital Humanities)

• Qualitative DH:
  • Datasets are typically small, curated, heavily annotated, flawless, specialized (‘smart data’)
  • Prototype: digital scholarly editions, e.g. Goethe, Faust Edition
• Quantitative DH:
  • Datasets are typically large, scraped, unannotated, with errors and biases, generic (‘big data’)
  • Prototype: Large Language Models trained on lots of text, e.g. ChatGPT

Reference: (Schöch 2013).

Third Way: Bigger Smarter Data

• In my opinion, there is a third way, which I have described already quite some tim ago as “bigger smarter data”.
• It’s not just a compromise: somewhat bigger, but still reasonably clean data;
• It’s an attempt to bring scale (quantitative) to the details (qualitative),
  
• Primarily using two technologies:
  • text mining or machine learning (for information extraction)
  • and linked open data (for data modeling and querying)

Reference: (Schöch 2013).

Background: What is Machine Learning?

• Fundamentally, ML involves detecting relations between features and labels
  • Features that we can observe in data
  • Labels, classes, or values that are relevant to our research
• We use this approach primarily for information retrieval
  • We start from a text collection
  • We may annotate part of the data
  • Then train a new model, or use an existing model
  • Evaluate the performance of the model on the annotated data
  • And then derive labels, classes, values from the unannotated text

Background: What is Linked Open Data?

• How does “Linked Open Data” actually work?
• There are subjects, predicates and objects
• Subjects are entities such as authors or works
• Objects are other entities, but also topics, locations, etc.
• Predicates link subject and object, like a verb
• This results in very simple statements
• The potential of LOD only unfolds with a large number of simple statements
• This seemingly simple structure is very powerful, because it is so generic
• To illustrate this power, one could liken it to chess: simple rules with lots of possibilities
• All the information we identify, using text and data mining, from our texts, is expressed in this way.
• A knowledge graph is created, which becomes part of the so-called “semantic web”

Source: Wikidata and datajournalism.com/read/longreads/the-promise-of-wikidata.

Linked Open Data: Multilingualism

• In LOD, all information is modeled semantically
• That means, each “unit of meaning” as an identifier
• And it has a lable and a description
• More precisely: many lanels and descriptions, in many languages
• So LOD and Wikidata are inherently multilingual
• That’s a big asset in a global scholarly world

Background: What is Literary History?

• Goals of literary history
  • Collecting and documenting knowledge of literary history
  • Providing explanations for the development of literature
• Organizational principles
  • Nations, periods, movements/currents, genres
  • Authors and works: themes, forms, relationships
  • Similarities and differences, continuities and change
• Functions
  • Describe and document literary history
  • Explain literary development

Literary History in Linked Open Data

• Building blocks
  • Subjects, including persons (author, etc.) and works (primary text, scholarly literature, etc.)
  • Objects, including works, but also themes, locations, protagonists, literary genre, etc.
  • Predicates, as required, including: author_of, about, sameAs etc.
    
  • Qualifications, e.g: Source (with type, date, URL)
• Some exemplary statement types
  • Bibliographic: [person] author_of [work]
  • Content-based: [work] about [theme]
  • Formal: [work] narrative_form [type]
  • and many more.

Wikidata for Literary History

• Idea: Create a “Wikidata for the history of literature”
  • Literary history information system
  • LOD-based, with explorative interface and SPARQL endpoint
  • Approach of an “atomization” of the historical knowledge
  • Linking with other knowledge systems (taxonomies, standard data, knowledge bases)
  • Key values: human and machine readable, open, collaborative, multilingual
• Compared to Wikidata
  • Focused on one domain (French novel, 1750-1800)
  • Better coverage / higher density of information for this domain
  • Development of a systematic ontology
  • much smaller: 300k vs. 1.5 billion statements

• Digital: algorithmic methods; available online; readable by humans and machines
• Multilingual: through semantic modeling
• Collaborative: necessarily interdisciplinary; networked with other resources
• Open: freely available software; all data and resources freely available; also already in the process

The project ‘Mining and Modeling Text’

• Mining and Modeling text brought literary history, machine learning and linked open data together.
• Mining and Modeling was a project funded by the regional government from 2019 to 2023
• This was an interdisciplinary project involving computer science, literary studies, legal studies, computational linguistics, and digital humanities
• The key goal of this project was to use computational methods to build a knowledge base about the French novel of the Enlightenment
• But more generally, also, to experiment with the idea of “Linked Open Data for Literary History”
• The MiMoText project has pursued this goal of collecting literary-historical knowledge, making it machine-readable and networking it using LOD
• We use three different sources of information for this purpose
  • Bibliographic metadata; in particular the “Bibliographie du genre romanesque francais, 1751-1800”, by Martin, Mylne and Frautschi
  • Knowledge from literary history, especially overview chapters from handbooks
  • And characteristics of primary texts, in our case based on a corpus of 200 French novels from the period 1751-1800
• Work objectives
  • Automatically extract relevant information from these sources
  • Model this information as LOD and link them together as much as possible
  • Analyze this information to learn more about the literature of the period, but also about literary historiography.
  • This also means transforming heterogeneous sources into a homogeneous data set
  • It also means explicitly modeling everything, and this is where controlled vocabularies, taxonomies, ontologies and their implementation and use come into play.

More information: (Schöch et al. 2022) and mimotext.uni-trier.de

Mining: Information Extraction

Pillar 1: Bibliographie du genre romanesque français

Reference: (Martin, Mylne, and Frautschi 1977).

Pillar 2: primary literature (novels)

• Corpus of 200 French novels (1750-1800)
• Encoding: in XML-TEI, with metadata, according to ELTeC schema
• Methods of analysis: Topic modeling, NER, stylometry, etc.

Collection of Eighteenth-Century French novels (1750-1800), ed. Julia Röttgermann, github.com/MiMoText/roman18. See also (Röttgermann 2024).

Pillar 3: Scholarly Literature

• Annotation guidelines => Manual annotations (using INCEpTION)
• Linking of INCEpTION with MiMoTextBase and Wikidata => disambiguation
• Creation of statements about authors and works (genres, themes, etc.)
• Machine Learning based on the annotated training data

Modeling: Data Modeling

Modular Data Model

• Module 1: Theme
• Module 2: Space
• Module 3: Narrative form
• Module 4: Literary work
• Module 5: Author
• Module 6: Mapping
• Module 7: Referencing
• Module 8: Versioning & publication
• Module 9: Terminology
• Module 10: Bibliography
• Module 11: Scholarly literature

Repository: github.com/MiMoText/ontology.

Example: The module on themes

Example: The module on narrative location

Meta-Statements

Linking with Wikidata for ‘federated queries’

Result: Queryable Database

The MiMoTextBase

• Because all statements are available in our Wikibase instance, we can now also search and query this knowledge network
• I will first show a few examples from the explorative view, similar to a wiki.
• Author: Not a lot of information, but: “exact_match” (!) => Wikidata
• Works: much more information, both as short texts and formalized and standardized according to the data model
• This information comes from various sources, so there are also meta-statements
• You can also search here, but not in a very precise way => SPARQL:::

See: data.mimotext.uni-trier.de

SPARQL endpoint

• The Wikibase Query Service allows very flexible queries
• There are also a lot of visualization options that can be used directly without further development work
• Our Wikibase instance now contains well over 300,000 statements, so the possibilities are quite broad.
• However, you need to familiarize yourself with the query language.
• That’s why we also offer a tutorial and lots of sample queries:::

SPARQL = SPARQL Protocol and RDF Query Language
See: query.mimotext.uni-trier.de

MiMoText Base: Query for themes in novels

• Every piece of information we capture is identified using authority data
• When using Wikidata, this means labels and definitions are available in multiple languages
• For our own taxonomies, we have focused on French, German and English
• When performing queries, we can mobilise this multilingual potential
• The above is a query for the dominant themes in the 200 novels of our corpus;
• We can easily switch between languages for labeling the graph

Live SPARQL query: query.mimotext.uni-trier.de

Some sample queries: simple queries

• List of novels with information from BGRF
• Number of works per author (first 25)
• Themes of novels, in French and in English

Example queries: visualizations

• Number of novels per year
• Narrative forms over time (decades)
• Book history: print formats over time (5 years)

Sample queries: networked and federated

• Link with catalogue data from French National Library (using BNF id)
• Narrative locations of novels (map)
• Authors by birth year, with portrait)
• Alternative author names from Wikidata infobox
• Network of influences between authors (using ‘influenced by’)
• Querying MiMoText from Wikidata (it works both ways)
• Novels and basic information, from Wikidata

Sample queries: comparative queries

• Themes from topic modeling compared to themes in BGRF
• Themes from BGRF vs. Topic Modeling (in one query)

Conclusion

Opportunities & challenges

• Opportunities
  • Linking heterogeneous data from different types of sources
  • Modeling, collecting and comparing contradictory statements
    
  • Transparency in knowledge production (sources)
• Challenges
  • Lack of consensus on relevant statement types in the discipline
  • Complexity reduction (triple structure)
  • Interoperability (tension ‘Wikiverse’ vs. OWL standard)

Lessons Learned

• Federated queries
  • Central element of the LOD vision
  • => Making it happen is not trivial (data model, infrastructure)
• Modeling meta statements
  • Very important: perspectives / statements, not facts
    
  • => Very different approaches in different technical contexts
• Exchange across communities
  • Literary Studies vs. Digital Humanities vs. Wikiverse
  • => is essential but needs more development
• There is still so much to do!
  • => We are continuing this effort in a new project called
    ‘Linked Open Data in the Humanities’ (LODinG)

Many thanks for your kind attention

Further resources

• Tutorial: https://docs.mimotext.uni-trier.de
• SPARQL endpoint: https://query.mimotext.uni-trier.de
• MiMoTextBase: https://data.mimotext.uni-trier.de
• MiMoText Ontology: https://github.com/MiMoText/ontology
• Reference publication: ‘Smart Modeling for Digital Literary History’
• Overview of visuals: mimotext.github.io/MiMoTextBase_Tutorial/visualizations.html

References

Martin, Angus, Vivienne G. Mylne, and Richard Frautschi. 1977. Bibliographie du genre romanesque français, 1751-1800. Mandell.

Röttgermann, Julia. 2024. “The Collection of Eighteenth-Century French Novels 1751-1800.” Journal of Open Humanities Data 10 (1): 31. https://doi.org/10.5334/johd.201.

Schöch, Christof. 2013. “Big? Smart? Clean? Messy? Data in the Digital Humanities.” Journal of Digital Humanities 2 (3): 1–19. https://journalofdigitalhumanities.org/2-3/big-smart-clean-messy-data-in-the-humanities/.

Schöch, Christof, Maria Hinzmann, Julia Röttgermann, Katharina Dietz, and Anne Klee. 2022. “Smart Modelling for Literary History.” International Journal of Humanities and Arts Computing 16 (1): 78–93. https://doi.org/10.3366/ijhac.2022.0278.