The large-scale analysis of historical events data makes it possible to trace key points of cultural and social exchange in history. There has been research focused on facilitating the integration and interpretation of events from heterogeneous sources (such as memoirs, books, and biographies) mainly considering events as a sequence of spatiotemporal objects. However, exploring and discovering new connections (e.g., collaborations, interactions) between people does require characterising those events with dimensions that are relevant to the scholarly enquiry such as the actual participants and nature of the event. This paper describes the concept of historical meetup to represent the encounters (for instance, collaborations, exchanges, links) between personalities of European history and formalise its constituent parts as an ontology. Furthermore, we report on preliminary work undertaken to generate a Knowledge Graph of historical meetups extracted from encyclopedic sources, i.e. biographies collected from Wikipedia. We discuss our results and illustrate the challenges of extracting such type of knowledge from biographical sources. The current experimental setting explores historical meetups in the European musical culture between 1800 and 1945. Our work sketches the basis for applying event knowledge graphs to cultural and social history research, providing support for the analysis, and exchange of ideas and practices.
The scholarly study of cultural and social heritage is a dynamic flow of experiences, leaving
heterogeneous traces dificult to capture, connect, access, interpret, and valorise. Social network
analysis (SNA) has gathered the attention of researchers across sociology and computer science
and can be a useful tool for cultural heritage research. By mapping out the connections, facts,
and interactions between individuals, SNA can support the exploration and understanding of
experiences, collaboration patterns and relationships, casting new light on aspects of cultural and
social exchange [
However, the interpretation and analysis of historical encounters using documentary sources
can be a complex task. For example, sources are usually heterogeneous and complex datasets
nEvelop-O
[
In this paper, we introduce the concept of historical meetups to describe historical encounters, for instance, collaborations and exchanges between personalities in European history. The MEETUPS Ontology models the constituent elements of a historical meetup, including entities such as people, places, temporal relations and types of events. The ontology aims to support researchers in performing a thorough analysis of social networks and discovering potential new links, therefore, addressing challenges in terms of interpretation and analysis complexity. The design rationale of the ontology was motivated by considering the knowledge requirements of scholars exploring social-historical networks and formalised as Competency Questions (CQs) in Section 4. Furthermore, MEETUPS Ontology is used as the framework to create a Knowledge Graph (KG) of historical meetups extracted from encyclopedic sources, for instance, Wikipedia biographies. The experimental setting explores social interactions in the European music world between 1800 and 1945. We develop our work in the context of the MEETUPS Pilot1 part of the Polifonia project. Our main contributions are: • The MEETUPS Ontology2 for representing historical encounters between personalities in
European history. • The preliminary work that implements a knowledge extraction pipeline3 for building a
KG of historical meetups from biographic sources, specifically Wikipedia. • The first version of the MEETUPS Knowledge Graph 4.
• A set of SPARQL queries for validation of the MEETUPS Ontology.
The rest of the paper is organised as follows. Section 2, presents a review of relevant research on the construction of KG for cultural heritage and SNA of historical meetups crafted from biographies. Section 3 details the methodology used to build the MEETUPS Ontology. Section 4 gives an account of the knowledge requirements and the development process for the ontology. Section 5 presents the evaluation of the ontology. Section 6 describes preliminary work to build a KG of historical meetups, and Section 7 summarises the challenges found in building the KG. Finally, Section 8 presents the concluding remarks and future work.
1MEETUPS Pilot: https://polifonia-project.eu/pilots/meetups/ 2MEETUPS Ontology repository: https://github.com/polifonia-project/meetups-ontology 3Knowledge extraction pipeline repository: https://github.com/polifonia-project/meetups_pilot 4MEETUPS Knowledge Graph repository: https://github.com/polifonia-project/meetups-knowledge-graph
The collection of data about historical encounters makes it possible to trace key points of
cultural exchange, dissemination and interesting connections between people, organisations
and other entities. However, documentary sources (e.g., books, biographies, memoirs) are often
unstructured and dificult to analyse, making it challenging for researchers to gain insights into
the lives of individuals and the historical contexts in which they collaborated. The construction
of knowledge graphs to represent structured data has the potential to shed light on the vast
documentary resources (physical and digital) and provide an overview of events that is both
comprehensive and easy to understand for scholars in the cultural heritage domain [
Initial work in cultural heritage and social network analysis focused on using Linked Data
(LD) and Knowledge Graphs (KG) to study general aspects of artist interaction. For instance, in
the musical domain, research explored how certain musical pieces influenced musical genre
development and aspects of music composition [
By drawing on the concept of historical events, current research expands on the development
of ontological models and KGs for implementing event-centric approaches. For instance, work
by [
In contrast to these studies, our approach focuses on modelling events from the perspective of
musical exchanges, including various entities, to bring a comprehensive view of meetings. Our
approach aims to design a whole process to extract data and build a knowledge base of historical
meetups. Unlike [
In what follows, we describe the methodology used to develop the MEETUPS Ontology and
present the model for representing historical meetups. We follow ontology engineering good
practices and methodologies [
A detailed account of the construction process for the MEETUPS Ontology is given in the following section.
This section is divided into two parts. The first part introduces the process for identifying and capturing knowledge requirements. The second part is dedicated to defining the classes, relationships and ontology construction.
First, we capture the knowledge requirements regarding support for historians and teachers in the exploration and visualisation of encounters between personalities in European musical history. The knowledge requirements were collected as part of collaborative work between scholars and researchers from the music department at The Open University and the Knowledge Media Institute (KMi); this work is part of the Polifonia project and the MEETUPS pilot.
Three Persona stories5 were developed to exemplify the knowledge requirements from the 5Persona stories: https://github.com/polifonia-project/stories perspective of target communities and scholarly users. The stories describe a series of scenarios regarding users’ scholarly activities, their data needs and interest in terms of music and cultural heritage, for instance, their requirements to analyse interactions between music personalities and the discovery of patterns for social exchange. The predominant Persona story for our experimental works is Ortenz, selected by music scholars. In what follows, we give a brief description of the Personas.
• Ortenz: She is a research fellow at the Music Department. Her background is in literature and art history. Ortenz would like to have a system for visualising events (meetings of composers and musicians) in time and space in order to track musicians’ careers, their overlap and intersections, gather trends in time and space and make emerging patterns of knowledge transmission. • David: He is a professor and music historian working in the music department of a university. David is interested in understanding the social history of music, e.g. who were the musicians, who was the audience, and how a particular musical environment relate to the wider musical environment. • Sophia: She is a musicologist and a practising musician. Sophia is interested in understanding the social-historical reasons behind how music was created and how it sounds.
We provide typical sentences from the historical domain that exemplify the information about personalities and events of interest for scholars. In this case, we take textual excerpts from Edward Elgar’s6 biography:
These examples are interesting for several reasons: (a) they contain details of events that influenced Elgard’s work: he attended other personalities’ performances (e.g., Saint-Saëns , he was inspired by contemporaneous personalities (e.g., Mary Lygon); (2) it contains explicit time references such as 1902 and 1880 which serve as guidance to search for other musicians that were working at the same time; (3) direct connections to other music personalities with whom he interacted; (4) and places he visited. For example, it is well accepted that European musicians influenced Elgard’s work; however, questions such as was he influenced by their surroundings (e.g., places he lived in, people he met, and shows he attended, what type of events)? Was his work part of emerging trends or patterns, when this happened? who were the contemporaneous emerging personalities? are of interest to music scholars.
Based on the description of the Personas’ stories, a list of CQs was generated. The list of relevant CQs is enumerated below:
6Edward Elgar biography, source Wikipedia: https://en.wikipedia.org/wiki/Edward_Elgar
Sophia Ortenz Ortenz Ortenz Ortenz Ortenz Ortenz Ortenz Ortenz Ortenz Ortenz
From this collaborative requirements acquisition exercise, it is clear that researchers expect to explore encounters that could potentially reveal unexpected connections and relationships. There is also a need for tools that support the analysis of social interactions from the perspective of ”events”, including layers such as time, places and participants. Therefore, the concept of historical meetups was introduced to describe the study of intersections and links among personalities from the perspective of the event and taking into account other dimensions, such as participants, place and time expressions.
The meetup type describes the reason/purpose for an encounter. According to scholars and researchers of the Music Department at The Open University, meetups can be grouped as business and career (e.g., contract, retirement), personal life (e.g., born, divorce, family), coincidence (e.g., meet, find, discover), education (e.g., learn, teach, conservatoire), public celebration or music-making (e.g., play, produce). Other dimensions of analysis include the participants involved, for instance, the person that is the subject of interest and the people interacting in the encounter, the place where it took place (e.g., city, country, venue), and the date when it took place. There could also be events that do not contain all the entities described previously.
From this analysis, we identify the vocabulary, relationships, and properties of the entities of interest. Therefore, this section is dedicated to describing the core components of the MEETUPS
Ontology. Furthermore, we explain how we incorporated established ontologies such as Time
Ontology [
The preferred prefix for MEETUPS Ontology is meetups: and the namespace for all MEETUPS terms is https://w3id.org/polifonia/ontology/meetups-ontology#. The core class of MEETUPS ontology is meetups:Meetup (see Figure 1), which is the representation of a historical meetup between people in the musical world. The main classes of the ontology are Purpose, Participant, Place and TimeExpression.
A historical meetup is defined by the purpose of the encounter, a meetups:Purpose class, which could be classified into one of the types of encounters: :BusinessCareer, :PersonalLife, :Coincidence, :Education, :PublicCelebration or :MusicMaking. A meetup involves at least one person (a meetups:Subject) and one or more participants represented by the class meetups: Participant that could take a specific role ( meetups:hasRole) in the musical encounter (meetups:Role). Historical meetups typically include a reference to the place (meetups:Place) where people or the activity developed and the date (meetups:TimeExpression) that indicates when this encounter took place.
Moreover, the ontology design includes classes for recording the provenance of the sources (see Figure 2). As described previously, typically data can be collected from encyclopedic resources such as biographies, and open public web pages; these sources are represented with the class meetups:Sources. At the same time, a source can hasProvider, represented by meetups:Provider class; for example, in our experiments, we use Wikipedia.
As described in Section 1, this work aims to represent historical encounters between personalities in European history. We built the MEETUPS Ontology following the list of knowledge requirements collected in Section 4. We use a sample of randomly selected biographies to build the KG and provide data for evaluation.
We used Protégé and the Hermit reasoner to check the formal consistency of the ontology. In order to evaluate the expressivity of the ontology, we grouped the Competency Questions (CQs) listed in Section 4 according to the most relevant type of entity it questions about and then encoded them into SPARQL queries7.
CQs (1 to 5) focus on place entities. These CQs inquire particularly about the locations where people met or visited. We take the example of the English composer Edward Elgar and build two queries (a) one that retrieves all the places he visited during his life, (b) one that retrieves all the places where he had an encounter with Adolf Pollitzer. As shown in Figures 3a and 3b, the ontology support such queries.
(a) Places visited by a given artist (b) Places linked to two artist encounters Figure 3: SPARQL queries for CQs focused on place entities
CQs (6 and 7) focus on participant entities. These competency questions refer to finding other contemporary artists and what were their social networks. We follow the example of Edward Elgar and built two queries (a) one that retrieves all contemporaneous artists given a specific date (see Figure 4a) and (b) a query that retrieves all people the artist met during their life and their role (see Figure 4b).
CQs (8 to 9) focus on time expression entities. These competency questions inquire about the time when artists met. We exemplify how the ontology answers this question by 7All available in the repository.
(a) People working at the same time Figure 4: SPARQL queries for CQs focus on participant (b) People met during their life building a query (see Figure 5a) that retrieves the date specific personalities met.
CQs (10 to 18) focus on purpose/events entities. Finally, these CQs inquired about what was the main purpose of the meeting. For instance, CQ 11 asks for the places where an artist performed (e.g., the meetup type is :MusicMaking), or CQ 12 questions where they live (PersonalLife), questions that can be answered thanks to the list of categories of events gathered as part of the knowledge requirements task (see Section 4). In the example (see Figure 5b), we query the list of meetups and their type.
(a) Date at which personalities met Figure 5: SPARQL queries for CQs 10 to 18 (b) List of events by type
This section describes the knowledge extraction pipeline developed to build the MEETUPS Knowledge Graph (see Figure 6). As described in Section 1 our work focuses on facilitating the discovery of connections between personalities of European history. Therefore, our approach is focused on identifying the entities that are part of a historical meetup: people, places, time expressions and meetup type. The entity recognition and disambiguation process is performed at the sentence level. We start by collecting data that provides an overview of music personalities’ lives, such as biographies in text format. Next, we process the corpus to link entities such as places and people, extracting time expressions and classifying them according to meetup types. The output of this step is a bag of entities at the sentence level, grouped by paragraphs and sections. Once the entities are extracted, we run the process to find encounters within the text, at sentence level or on a larger chunk of text using coreference resolution. The last step is dedicated to building the knowledge graph of historical meetups using as a framework the MEETUPS Ontology and the data output from the previous step. In what follows, we describe in detail the steps taken towards constructing the KG.
Data collection. The first step of the pipeline is dedicated to the collection of data that supports music historians’ exploration of encounters between people in the musical world in Europe between the 1800s and 1945. Therefore, we rely on documentary sources such as biographies and select an open-access database such as Wikipedia as the primary data input. To obtain the list of targeted personalities, we built and queried the DBpedia SPARQL endpoint. We filtered biographies by date of birth in order to reflect our targeted sample and gathered a total of 33,309 biographies. Each biography was collected in text format from Wikipedia, using the identifier from DBpedia. In order to make experiments manageable, we randomly sampled a total of 1000 biographies that will be inspected in the following steps.
Cleaning & Segmentation. In the previous step, the biographies were collected in text format; therefore, the second step is dedicated to preparing the dataset to execute the NamedEntity Recognition and Disambiguation (NERD) tasks. First, we clean the text by removing line breaks, special characters and empty spaces. In order to facilitate the location and classification of the entities, the text is organised in sentences, grouped by paragraphs and sections. The ifnal output is a corpus of biographies, each one containing a set of sentences identifiable by its position within paragraphs and sections.
Entity recognition. As explained previously, the main objective is to explore encounters by
characterising them with dimensions that are relevant to scholars, specifically: people, places,
temporal expressions and types of encounters. Therefore, we focus on locating and identifying
the constituent parts of a historical meetup within sentences. In what follows, we detail the
tasks developed to recognise each type of entity:
• Named-entity recognition and disambiguation (NERD) for people and place entities. People
and places are two main elements that characterise a historical meetup. As listed in Table 1,
knowing who was involved and where encounters happened is required. Therefore, this
task is focused on determining people or places mentioned in the corpus and specifying
them uniquely using a targeted knowledge base. We make use of DBpedia Spotlight8
tool to automatically annotate mentions of people and places linking them to DBpedia
resources.
8DBpedia-Spotlight: https://www.dbpedia-spotlight.org/
• Identification of temporal expressions. Another important requirement has to do with
the date or moment in time when a particular encounter happened. To extract this
information, this task uses a rule-based Time Expression recognition tagger based on
research by [
Identification of meetup themes. The main element of a historical meetup is the theme, in
other words, the topic/subject of the encounter. This task uses Machine Learning techniques and
performs a semi-supervised classification process to annotate the meetup type (see Section 4.1)
of each sentence in the corpus. In what follows, we describe the steps of the theme identification
process in detail:
• Building the training dataset. In order to create the training dataset, we use a distant
supervision approach. First, we built a curated list of keywords for each one of the meetup
types; the list was developed with the advice of domain experts. Next, we randomly
select a sample of sentences and annotate them automatically with one of the classes
(meetup types) if it includes one of the keywords in the list. The final output is a dataset
of sentences (a total of 16140) annotated according to one of the meetup types (see Table 2
for statistics).
• Classification task - training and testing of Machine Learning (ML) algorithms. In this step,
we use the dataset built previously and implement a supervised classification approach.
As shown in Table 2, the distribution of sentences by class is unequal; therefore, the
ifrst task is building a balanced dataset. We take as reference the lowest number of
sentences per group, in this case, the class Education with 629 sentences. For each class,
we select randomly 629 annotated sentences to ensure each class has the same number of
annotations. The new training dataset has a total of 3774 annotated sentences. Then, we
prepare the dataset for the training process; we use SBERT to capture the context of the
sentences [
The final output is a sentence-level bag of semantic entities: meetup types, people, places and temporal expressions. We aim to use this new corpus to build a knowledge graph of historical meetups.
9SynTime software: https://github.com/zhongxiaoshi/syntime
Knowledge Graph construction. In this last step, we focus our eforts on building a knowledge graph of historical MEETUPS, using the entities found at a sentence level. Crucially, we aim to identify missing elements at the sentence level and annotate them for resolution in the next version of the KG, for instance, implementing a coreference resolution task to determine if missing elements are located in previous sentences.
We built the KG following the MEETUPS Ontology model. The entities found at the sentence
level represent a meetup candidate. The KG also stores provenance data (e.g., source, type of
source, location identifier in the text). The KG was constructed from CSV files resulting from
the process described so far, using SPARQL Anything [
The MEETUPS KG contains data from 1000 artists’ biographies. As summarised in Table 3, around 74445 historical meetups, there are 51425 mentions of people involved in diferent encounters. So far, the historical meetups gather around 5595 places and 79838 time expressions.
The evaluation of the MEETUPS knowledge graph has been made incrementally to ensure the data extraction process complies with the knowledge requirements (see Section 4). Here we present the input data validation using the first version of the KG. The data covers people, place, time expressions and meetup types. All these entities were extracted at the sentence level. We designed and executed a short survey as follows: • Select randomly the sentences from musicians’ biographies for evaluation. – The evaluation dataset was built from a random sample of 12 biographies from the total analysed in the first iteration (1000 biographies). – Due to time and resource constraints regarding the annotation and evaluation of the sentences, we selected one paragraph and its corresponding sentences from each biography. Some paragraphs contained only one sentence, others up to seven sentences, reaching a total of 40 sentences. Each sentence listed the total entities resulting from the knowledge extraction pipeline. • For the data collection process: – For each sentence, we asked participants to analyse the correctness of the automatically identified entities in terms of the number of sentences identified as meetups, and the number of people, places, time expressions and meetup types correctly extracted. – For each sentence, we asked participants to indicate the number of missing entities not identified by the automatic annotation process.
From the forty sentences, 34 (85%) were identified as potential meetups/encounters. Table 4 displays in detail the number of sentences that mention a type of entity. Of the 34 sentences identified as a meetup, nine sentences (26%) provide information about all four types of entities. Twenty-five sentences (74%) lack one or two of any of the types of entities.
We move now to analysing the entities extracted at the sentence level. First, regarding the completeness of sentences to represent a historical meetup: • Results show that 26% of sentences (1 out of 4 sentences) are comprehensive (including all four types of entities), and 74% mention at least one out of the four types of entities. Therefore, it is possible that a historical meetup is not fully identifiable from one sentence. For example, the sentence: “Born Al Albertini in Chester, Pennsylvania, United States, he went to South Philadelphia High School.” does not include a time entity but still provides data to answer the CQs. Future work regarding the completeness of historical meetups should explore cases of missing entities, such as searching for information in preceding sentences at paragraph and section levels.
Second, regarding the quality of the extracted entities: • As mentioned in the previous point, sentences representing historical meetups include one or all four types of entities. Of the total of sentences analysed, 62% of them mention people and 38% of sentences do not (see Table 5). From the sentences that refer to people, the knowledge extraction process correctly identifies them in 62% of the sentences and missed people entities in 19% of sentences. People were incorrectly and partially identified in 19% of the sentences.
– The cases in which people were incorrectly identified mainly pertain to homonyms names. For example, people’s names are used to name prizes, and festivals, among other cases. Further validation includes querying DBpedia/Wikidata to determine if the entity is a person (for example, has a date of birth) and the entity type (for example, if it refers to a person or musician). – In cases where people entities were partially identified, we found most cases referred to entities named in a previous sentence or at the beginning of the paragraph. The next steps include exploring neighbouring sentences to identify such entities and implementing a coreference resolution task. • From sentences that mention places, the knowledge extraction process correctly identifies place entities in 87% of the sentences, and only 9% of sentences were missed. Place entities were partially and incorrectly identified in 4% of the sentences. For the few cases in which places were missed, incorrectly and partially identified, the analysis shows the definition of places should be expanded. For instance, ’Conservatoire de Paris’ can be identified as being in ’Paris’; this entity is of type dbo:Campus not dbo:Place. Future work should consider the level of detail required to recognise place entities. • In both cases, people and place entities, a few instances were missing in DBpedia; for these cases it is essential to consider other databases such as Wikidata to link resources. • From sentences that include references to time, the knowledge extraction process correctly identifies them in 42% of the sentences and 58% of sentences were missing. The analysis shows that in most cases, time expressions such as years (e.g., in 1923, 1899-1900), relative expressions (e.g., mid-eighties, late 1960) or indirect references (e.g., two months later, last year) are missing. The time expression identification process should expand the heuristic rules in the next iteration. • Regarding the meetup type identification, all sentences are annotated with one of the classes identified in Section 4.1. The knowledge extraction process correctly identifies the correct type in 74% of the sentences. The analysis shows that in most cases, the meetup type is partially correct or incorrect when one sentence refers to more than one event in the same sentence. The time expression identification process should consider these cases in the next iteration.
In this paper, we presented the MEETUPS Ontology, a model that formally characterises the encounters between personalities of European history. We introduced the concept of the historical meetup to describe such encounters, for instance, collaborations and knowledge exchanges between people. The ontology describes the elements of a meetup, including entities such as participants, locations, time expressions and meetup types. We presented our approach as an essential tool to address the knowledge requirements of musical scholars and cultural heritage researchers who face challenges when analysing artists’ social links and influences. Importantly, in collaboration with experts, we compiled a list of knowledge requirements expressed as Competency Questions (CQs). These CQs later served to validate the expressivity of the ontology.
Furthermore, we presented the preliminary work on developing the Knowledge Extraction pipeline for identifying the elements of historical meetups. The experiments were developed in the music cultural heritage domain context and used as data source biographies collected from Wikipedia. The execution of the pipeline produced the first version of the MEETUPS Knowledge Graph, crucially providing an overview of the main challenges to tackle to extract and represent the constituent elements of a historical meetup. Therefore, future work includes addressing the challenges summarised in Section 7 and the data quality evaluation of the knowledge graph.