Musical Meetups Knowledge Graph (MMKG): a collection of evidence for historical social network analysis

Knowledge Graphs (KGs) have emerged as a valuable tool for supporting humanities scholars and cultural heritage organisations. In this resource paper, we present the Musical Meetups Knowledge Graph (MMKG), a collection of evidence of historical collaborations between personalities relevant to the music history domain. We illustrate how we built the KG with a hybrid methodology that, combining knowledge engineering with natural language processing, including the use of Large Language Models (LLM), machine learning, and other techniques, identifies the constituent elements of a historical meetup. MMKG is a network of historical meetups extracted from ∼33k biographies collected from Wikipedia focused on European musical culture between 1800 and 1945. We discuss how, by providing a structured representation of social interactions, MMKG supports digital humanities applications and music historians’ research, teaching, and learning.

1. Musical MeetupsKnowledge Graph (MMKG):
a collection of evidence forhistorical social
network analysis
MAY 2024
ESWC 2024 – Resource Track
Alba Morales Tirado, Jason Carvalho, Marco Ratta, Chukwudi Uwasomba,
Paul Mulholland, Helen Barlow, Trevor Herbert and Enrico Daga

2. BACKGROUND
Leveraging the study of cultural and social
exchange through the study of musical encounters
Historical research, such as social music history, relies heavily on
information contained within historical documentary evidence
regarding the biographies offigures from history.
Some problems may arise:
•Find historicalencounters within documentary sources
can be a complex and intensivetask
•Building comprehensive storylines of people's
encountersandinteractions requires several dimensions:
spatial, temporal, and thematic
•Large and heterogeneous datasets. E.g., books,memoirs,
biographies (prosopography).

3. Modeling MMKG
We use as a guiding framework the Meetups Ontology [1]
MEETUPS ONTOLOGY
[1] Morales Tirado, A., Carvalho, J., Mulholland, P. and Daga, E., (2023). Musical Meetups: a Knowledge Graph approach for Historical
Social Network Analysis. In: CEUR Workshop Proceedings: SEMMES 2023: Semantic Methods for Events and Stories workshop ESWC 2023. ​
Historical meetup a formal definition:
Describes historical encounters, for instance,
collaborations and exchanges between
personalities in European history.
Main classes:
•mtp:Meetup, is derived from evidence within a
biography
•mtp:Participant and mtp:Location classes
represent mentions of at least one or more
participants and places, respectively.
•mtp:TimeExpression class is composed of start
time:hasBeginning and end time:hasEnd
•mtp:Purpose each meetup can be annotated with
more than one purpose

4. Knowledge Graph generation pipeline
The main objective is to support scholarly activities by providing tools that allow the exploration
and visualisation of encounters between personalities in European musical History that could
potentially reveal interesting or unexpected connections or relationships

5. ENTITY RECOGNITION
a) People and places​
•DBpedia Spotlight tool to
automatically annotate
mentions of entities. ​
•Linking to DBpedia resources
b) Temporal expressions​
•Our software uses NLTK Toolkit
and was developed using Python​
•We use a rule-based Time
Expression recognition tagger
Knowledge Graph generation pipeline
A hybrid pipeline to recognise and link the
entities that are part of a historical meetup
•33.309 biographies
collected in text format​
from Wikipedia
DATA COLLECTION
Participants
Example:​
“In his first trips abroad, Elgar
visited Paris in 1880 and
Leipzig in 1882. He heard
Saint-Saëns play the organ at
the Madeleine and attended
concerts by first-rate
orchestras.”​
Location
Time
expression
Elgar Edward​
Saint-Saëns​
Paris​
Leipzig​
1880​
1882​

6. Inclusion of LLM (in-context learning), resulting on
50% of such expressions being parsed
Knowledge Graph generation pipeline
ENTITY RECOGNITION
c) Temporal expressions normalisation
•Time as ranges, with a start and end time
point
•ISO8601 format (YYYY-MM-DD)
•Python libraries: dateutil and approx-dates
On average 65% of temporal expressions
are normalised.
•E.g.: “1880"
⚬Start date: 1880-01-01
⚬End date: 1880-12-31
35% corresponds to expressions such as
•“the next seven years” or
•“of the twentieth century”
•Output: date in ISO format
•Input:
⚬the context of the temporal expression
⚬the sentence where the expression was identified
⚬the subject’s name (biographee)
Prompt for normalisation

7. Meetup
type
Knowledge Graph generation pipeline
ENTITY RECOGNITION
Method 1
Initial Machine Learning approach
•A semi-supervised classification process
⚬Built a training dataset using a distant supervision
approach
⚬Training and testing of ML - MLP classifier
•Annotating each sentence in the corpus
and assigning them one of the meetup
types
Method 2
Using LLM to identify meetup themes
•Zero-shot learning approach​
•Input: text to analyse and the list of classes​
•Output: two relevant themes and the reason
•Business and career
•Personal life
•Coincidence​
•Education​
•Public celebration​
•Music making​
d) Identification of meetup themes
Classification results. Comparison ML and LLM Precision @ 1 and 2

8. •We use SPARQL Anything and design CONSTRUCT
query mappings, to materialise the triples
•Downloadable files and SPARQL Endpoint
•https://polifonia.kmi.open.ac.uk/meetups/sparql/
Knowledge Graph generation pipeline
HARMONISATION KG CONSTRUCTION
Coreference resolution
Identification of Historical Meetups
MMKG Statistics
•Identify implicit mentions, in the form of noun
phrases or pronouns
•Maximise the identification person or a
place entities
•Using the spaCy library coreferee
•An algorithm to identify historical meetups that
have at least one participant (plus any place,
temporal expressions and meetup type).
ParticipantsHistorical meetup:
“His only formal musical
training beyond piano and
violin lessons from local
teachers consisted of more
advanced violin studies
with Adolf Pollitzer, during
brief visits to London in
1877–78.”
Location
Time
expression
Elgar Edward​
Adolf
Pollitzer
London
1877-01-01
to
1878-12-31
Theme
Music
making
Ad: DEMO 27
PySPARQL-Anything Showcase

9. CQs focus on place dimensions CQs focus on purpose
Evaluation
ANSWERING COMPETENCY QUESTIONS
The objective is to evaluate whether the MMKG data meets the knowledge requirements
Queries and results obtained are available in the MMKG repository - queries folder
https://github.com/polifonia-project/meetups-knowledge-graph/
•What places did musician Z visit in his/her career?
•Where did musician X and performer Y meet?
Answer: a query to find the places German pianist
Clara Schumann and Joseph Joachim met
•Why did musician X and performer Y meet?
•What is the nature of the event (a celebration, a
festival, a private event, a performance, accidental)?
Answer: Clara and Joachim’s meetings mainly describe
“Music Making” theme.
For instance, “In October–November 1857, Schumann
and Joachim went on a recital tour to Dresden and
Leipzig.”

10. CQs focus on temporal expressions CQs focus on person dimension
Evaluation
ANSWERING COMPETENCY QUESTIONS
The objective is to evaluate whether the MMKG data meets the knowledge requirements
Queries and results obtained are available in the MMKG repository - queries folder
https://github.com/polifonia-project/meetups-knowledge-graph/
•When did musician X and performer Y meet?
•Did musician X and performer Y ever meet?
Answer: a query to extract textual evidence of the time C.
Schumann and Joachim met, available in the KG
•Who other musicians were working at the same time?
•What was the composer’s network?
Answer: a query to retrieve the list of people Clara
met, all part of the KG

11. Results
Feedback questionnaire from domain experts
•Surveyed to evaluate the value of MMKG for domain experts
•12 domain experts from the Music Department of The Open University participated
Findings of potential adoption
Importance of documenting the following dimensions such
as people, location, dates and themes
The usefulness of
knowledge graphs
in teaching music
history
Reported daily
engagement
91.7%
Find it useful for
teaching purposes
75 %
Engagement with
music-related
content

12. Conclusions & Future work
•Musical Meetups Knowledge Graphs: a knowledge graph of documentary
evidence of social interaction for supporting research in music history.
⚬Crucially enriched with participants, geolocation, time-indexed and
thematic annotations
⚬Novel hybrid methods for knowledge extraction that combine knowledge
engineering with techniques from traditional NLP and current LLM tools
•A web interface leveraging the potential of the MMKG has been published
and is under testing with domain experts
•Future work on linking Meetups mentioned in different biographies
•Future work on incorporating main historical events (context)
•Timeline linking to other existing KGs (e.g. EventKG, WikiData)
•We plan further research regarding
•network analyses (e.g. influence)
•Supporting researchers in formulating new hypotheses

13. THANK YOU
* This project has received funding from the European Union’s Horizon
2020research and innovation programme under grant agreement GA101004746
MAY 2024
Resource Availability
•SPARQL endpoint
https://polifonia.kmi.open.ac.uk/meetups/sparql/
•Repository
https://github.com/polifonia-project/meetups-knowledge-graph
•Meetups Web Interface
https://polifonia.kmi.open.ac.uk/meetups/
Meetups