The underlying idea of this paper is to represent life stories of people as Linked Data, extracted and aggregated from heterogenous distributed data sources, such as dictionaries of national biographies, museum collections, library databases, Wikipedia etc. (Hyvönen et al., 2018). Linked biographical data facilitates studying enriched individual life stories in biography research (Roberts, 2002) as well as in prosopography research on groups of people (Verboven et al., 2007;Keats-Rohan, 2007). This paper addresses the fundamental technical research question that has to be solved in this kind of work: how to model life stories, so that they can be enriched from heterogeneous data sources and interlinked with each other in a semantically interoperable way? Our research hypothesis is that a good choice for data modeling and harmonization is the event-based approach where a person's life is seen as a sequence of spatiotemporal, interlinked events from birth to death-a person may also be involved in prenatal and posthumous events. For example, metadata about a painting in a gallery actually means that there has been a painting event, and this could be included in the timeline of the artist's semantic biography. Event-based modeling and ontologies have already been found useful for harmonizing heterogeneous cultural heritage data. A most notable and widely used ontology for this is the CIDOC Conceptual Reference Model (CRM) 1 (Doerr, 2009), but there are also other models (Raimond and Abdallah, 2007;Scherp et al., 2009;Shaw, 2010;van Hage et al., 2011). A recurring problem in event based modeling is, however, that it is not necessarily clear what is an event, since many 1 http://cidoc-crm.org relations and roles (Kozaki et al., 2006) occur in time and space, too. For example: Are family relationships events, e.g., being the father of or being married to someone? Are professions events, such as being a president of a country, because holding an office occurs in time and space with an agent involved? For example, the concept of "bishop" would be useful in representing and querying biographical data, but what does being a bishop actually mean? Is there a class and instances of bishops, is being a bishop a property of a person or a role, or how does the concept relate to the event of holding a bishop's office? Obviously, being a bishop can be represented in different ways, but then harmonizing and querying of data about bishops becomes very difficult since the user cannot be sure in what alternative ways being a bishop is actually represented. On the other hand, we clearly need foundational ontological structures (Guarino and Welty, 2002) for representing pieces of heterogeneous knowledge in a systematic and unique way, but on the other hand, there is a need for simple conceptualizations and property structures for querying the data and representing the data for the human user. To address these problems, this paper introduces a data model, Bio CRM, for harmonizing, enriching, and using biographical linked data based on events. Bio CRM is an extension of CIDOC CRM to the biographical domain. This ISO standard was selected as the basis because it is the most widely used ontology standard for event-based modeling in museums and has been integrated with the Functional Requirements for Bibliographic Record (FRBR) family of modeling standards in libraries2 . Data from museums and libraries are essential in describing life stories.
In the following, major use cases for Bio CRM in biographical and prosopographical research are first listed. After this, the design principles and the actual data model are presented, with three online applications illustrating the use of the system. Finally, related work is discussed with a comparison of Bio CRM with other data models.
The aim of using the Bio CRM data model in our case studies is to facilitate using the prosopographical research method (Verboven et al., 2007, p. 47) that consists of two major steps. First, a target population of people is selected that share desired characteristics for solving the research question at hand. For example, our research question may be related to social networks of men who were born in Finland 1800-1900 and were artists. Second, the target group is analyzed further, and compared with other groups, in order to solve the research question. 1. Determine target groups Target groups can be found by data filtering with a human in the loop. For example, SPARQL SELECT can be used to create a tabular set of selected instances. In our case, faceted search is a promising option for filtering out target groups in a flexible and dynamic way. An interesting possibility for further research would be to try to do filtering automatically using knowledge discovery. Once a target group has been determined, specific working hypotheses and specific historical questions concerning the group can be formulated and analyses performed. 2. Prosopographical analysis Linked data and SPARQL querying provides many possibilities for analyzing target group data. For example, it is possible to analyze the structure and changing composition of the group in time and the changing roles of individuals or subgroups. In this case the result of a SPARQL SELECT or CONSTRUCT is analyzed further by specific algorithms or visualization tools, such as information graphics. Another option is to employ methods of network analysis methods and tools (Easley and Kleinberg, 2010;Hanneman and Riddle, 2005) and visualizations (Dadzie and Rowe, 2011;Kehrer and Hauser, 2013). In this case, for example, a SPARQL CONSTRUCT can be used for creating an RDF network based on the underlying data.
To aggregate, enrich, and link biographical data with related datasets the data must be made semantically interoperable, either by data alignments (using, e.g., Dublin Core and the dumb-down principle) or by data transformations into a harmonized form (Hyvönen, 2012). In our case we selected the data harmonization approach and the eventcentric CIDOC CRM ISO standard as the ontological basis, since biographies are based on life events. CIDOC CRM provides a common and extensible semantic framework for representing cultural heritage information, operating as a "semantic glue" for integration, mediation, and interchange of heterogeneous datasets from, e.g., museums, libraries, and archives. In our work, biographies are modeled as collections of CIDOC CRM events, where each event is characterized by the 1) actors involved, 2) place, 3) time, and 4) the event type. Bio CRM extends CIDOC CRM by introducing role-centric modeling. The reason for the extension is that while CIDOC CRM does include a mechanism for representing roles of participants in events, its encoding in RDF is complex and still in experimental phase (see Section 5 for further discussion). Bio CRM provides the general data model for biographical datasets. The individual datasets concerning different cultures, time periods, or collected by different researchers may introduce extensions for defining additional event and role types. The Linked Data approach enables connecting the biographies to contextualizing information, such as the space and time of biographical events, related people, historical events, publications, and paintings. The core design principles of the data model are:
• The model is a domain specific extension of CIDOC CRM, making it applicable to not only biographical data but to other Cultural Heritage (CH) data, too.
• The model makes a distinction between enduring unary roles of actors, their enduring binary relationships, and perduing events, where the participants can take different roles modeled as a role concept hierarchy.
• The model can be used as a basis for semantic data validation and enrichment by reasoning.
• The enriched data conforming to Bio CRM is targeted to be used by SPARQL queries in a flexible ways using a hierarchy of roles in which participants can be involved in events.
Bio CRM makes a clear distinction between person's attributes, relations between people, and events in which people participate in different roles.
• Attributes are properties of a person that are assumed to characterize her independently of time and space. For example, place and time of birth can be modeled as attributes.
• Relations are established between people and are assumed to characterize them independently of time and space. For example, father-of is such a relation. Relations can, however, have time and space as qualifiers, e.g., student-of. For example, Ferdinand Bol (1616-1680) was a student of Rembrandt in 1630-1641, starting his own studio in 1641, but can be characterized as a student-of Rembrandt in general. His years in Rembrandt's studio as a student can be represented as an event (see below), if needed.
• Events take place in time and space and involve participants in different roles, expressing the ways in which persons participate in events. For example, an officiant may participate in a certain baptism event.
The core classes and properties of Bio CRM are presented in Figure 1. Similary as for unary roles, the binary role classification can be extended in individual datasets. The individual events of biographies are represented as subclasses of bioc:Event that is a subclass of cidoc:E5 Event inheriting its properties. From a semantic viewpoint, events are described especially in terms of
• the time of the event (cidoc:P4 has time-span),
• place of the event (cidoc:P7 took place at),
• actors that participated in it (cidoc:P11 had participant),
• other resources involved (cidoc:P12 occurred in the presence of).
Time and place properties refer directly to time spans and instances of places, respectively. The values for participating actors and other resources are instances of role classes. An actor role associates an actor with a role, making it possible for a person to participate in events in different roles that can also be qualified in terms of additional properties.
Similarly to actors, physical objects and immaterial things can be involved in an event in specific roles.
Events can be used for qualifying a unary (e.g., an occupation) or a binary relation further, i.e., in such cases an instance of bioc:Event has to be created. As an example, Figure 4 validating data, i.e., to see if the events have participants in incompatible roles. It is recommended that each event class, say Baptism, has a corresponding class of allowed roles, say Baptism Actor Role. Its subclasses are roles whose instances can be used in filling the roles. In this way, the data annotator can be guided to use only the correct roles, and the new role class can be used for finding resources in roles when querying. The role hierarchy facilitates sharing roles between events and modifying their role structure easily by just editing the role hierarchy. This is more flexible than, e.g., changing property names, if roles were represented using different properties. The following SPARQL query is an example for finding all "bishops" in a dataset. Note that because of the chosen role modeling approach, the query returns both the bishops as unary roles (occupation) and acting bishops in specific events (e.g., a confirmation). The namespace prefix declarations are omitted from the query for brevity.
In the following, three case studies for using Bio CRM are presented.
Bio CRM was originally developed as a spin-off case study related to the database and web service Early Modern Letters Online (EMLO)3 . EMLO is a collaboratively populated union catalogue of sixteenth-, seventeenth-, and eighteenth-century letters, created by the Cultures of Knowledge project4 at the University of Oxford. It brings manuscript, print, and electronic resources together in one space, increasing access to and awareness of them, and allows disparate and connected correspondences to be crosssearched, combined, analyzed, and visualized. In addition to purely epistolary data, EMLO contains prosopographical information related to the people in the database, modeled as events and social relationships. Events cover activities that the people have participated in during their lives, such as birth and death, ecclesiastic and educational activities, creations of works, travels, and residences. The event metadata includes the event name, type, participants and their roles, time span, location, and source information. As a pilot Linked Data publication of the EMLO database (Tuominen et al., 2018), we have converted the prosopographical data into RDF format using CIDOC CRM for the event-based modeling and W3C's PROV model (Lebo et al., 2013) for representing the roles of participants in the events. As a next step, we propose to convert the data into a Bio CRM representation, and build the event and role hierarchies pertaining to the activity types stored in the database. The top levels of the event hierarchy of the EMLO database are the following ones: bioc:Event :Ecclesiastical_Event :Educational_Event :Political_Event :Professional_Event :Social_Status_Change
The class :Ecclesiastical Event can be divided further into subclasses with attached roles, such as:
• :Baptism -:Officiant, :Baptismal Candidate, :Godfather, :Godmother, :Religion
• :Confirmation -:Officiant, :Confirmation Candidate, :Religion 4.2. The register of the high school "Norssi" alumni Bio CRM has been applied in the study of transforming printed biographical registers into Linked Data and enriching their contents using Named Entity Linking. As a concrete case study, we have concentrated on the printed register "Norssit 1867--1992. Helsingin Norssin matrikkeli", a book of 708 pages, containing short bios of over 10 000 students and teachers of the prominent Finnish high school "Norssi", a training school of the University of Helsinki. The final application in use5 based on this case study is described in more detail in (Hyvönen et al., 2017), and the underlying data model is presented in (Leskinen et al., 2017).
Bio CRM has also been used in creating the first prototype demonstrator of the Semantic National Biography of Finland (Hyvönen et al., 2018), based on a collection of some 13 000 short biographies that were transformed into RDFformat and enriched using data linking to external datasets. Application of Bio CRM to prosopographical research in the Norssi and Semantic National Biography case studies is described in more detail in (Leskinen et al., 2018).
Bio CRM is to the best of our knowledge the first attempt to extend CIDOC CRM into the domain of biography and prosopography with additional subclasses, properties, and modeling guidelines. A major benefit of the model is the compatibility with cultural heritage data from museums, libraries, and archives represented using the same standard framework ontology. From a modeling perspective, this paper presented the idea of making a distinction among attributes, relations, and events, where entities participate in different roles in a qualified manner, not as entities themselves. The underlying rationale for this is to harmonize the knowledge representation with fewer categories and at the same time keep the model expressive and easy to use by SPARQL queries. If needed, the model can be extended with transformational rules, by which more expressive and foundational event structures can be transformed into simpler attribute and role structures, when needed and appropriate from an application perspective, and vice versa. Events are needed for harmonizing data for the machine but the human end user often conceptualizes the world in document-centric or other ways. Thus, additional representations of the same eventbased knowledge may be useful, especially if it is generated automatically by the system. Our experiences of applying a first version of Bio CRM to the three case studies suggest the the model is usable for practical purposes. Though formal evaluation of the model has not been conducted, the application of Bio CRM to data originating from different sources, in different formats, and covering different eras is an indication of the suitability of the model to act as a general harmonizing model for prosopographical data. In the spirit of design science methodology, Bio CRM was designed to solve the modeling needs of the prosopographical data of the Early Modern Letters Online (EMLO), which provides a rich classification of the events and associated roles of people in the early modern era. In the cases of high school "Norssi" alumni and Semantic National Biography, Bio CRM is used to model, e.g., the family relations and titles of the people (e.g., education, occupation). The idea of using semantically defined linked data for modeling and aggregating biographical and related data seems to be a promising approach for biography and prosopography. However, more work is needed in detailing out more precisely the class and property structures of the model in the general case-so far new classes and properties have been introduced based on the particular use cases and the general modeling principles presented in this paper. Biographical data has been studied by genealogists (e.g., (Event) GEDCOM6 ), CH organizations (e.g., the Getty ULAN7 ), and semantic web researchers (e.g., BIO ontology8 ). Semantic web event models include, e.g., Event Ontology (Raimond and Abdallah, 2007), LODE ontology (Shaw, 2010), SEM (van Hage et al., 2011), and Event-Model-F9 (Scherp et al., 2009). Also, Bibliographic Ontology (BIBO) (D'Arcus and Giasson, 2009) includes a model for events. For a more detailed comparison on event models, see (Scherp and Mezaris, 2014). A history ontology with map visualizations is presented in (Nagypal et al., 2005), and an ontology of historical events in (Hyvönen et al., 2007). Visualization using historical timelines is discussed, e.g., in (Jensen, 2003), and event extraction reviewed in (Hogenboom et al., 2011). PROSO (Zingoni, 2014) is a data model for presenting prosopographical data records. It has a strong focus on representing the provenance information of the records using factoids, and uses event-based modeling for stating the changes of a person (e.g., receiving a honorary ti-tle). Vocabularies and ontologies for representing personal relationships include the Standards for Networking Ancient Prosopographies (SNAP) project10 and the Relationship vocabulary (Davis, 2004) (Smith et al., 2015) and the related Relations Ontology (RO) 13 , inheres in to attach the role to a person, and realized in to attach the role to an event (and their inverse properties bearer of and realizes). It should be noted that the inclusion of the properties inheres in/bearer of and realizes/is realized in in BFO is unclear in the current version (BFO 2.0) 14 . Bio CRM has taken some inspiration from VIVO, and uses similar properties bioc:inheres in and bioc:bearer of, but retains compatibility with CIDOC CRM by using the property cidoc:P11 had participant to attach the role to the event. Schema.org 15 provides a class schema:Role to represent additional information of a relation or property.
It can be used to model the roles of participants in events. The instance of the schema:Role acts as an intermediary node between the event and participant, both of them attached with the "original" property, e.g., cidoc:P11 had participant. The model's strength is its simplicity, but the re-use of a property in such a way might be unintuitive. The model also introduces a new property schema:additionalType, instead of using the standard RDF property rdf:type.
W3C's PROV model uses qualified associations to specify the roles of the participants in events. The instance of the class prov:Association is attached to the role using the property prov:hadRole, to the agent using the property prov:agent, and to the event using the property prov:qualifiedAssociation. Thus, the standard CIDOC CRM event can be qualified with such an association, but it might be unintuitive for the user to represent the qualifier separately from the cidoc:P11 had participant relation. By design, PROV is meant for representing provenance information involved in producing a piece of data or thing; all biographical events are not such activities. W3C's Organization Ontology (Reynolds, 2014) is a core ontology for describing organizational structures. It includes a model for specifying membership roles people have in organizations by introducing the class org:Membership. Such a membership is associated with the role using the property org:role, to the agent with the property org:member, and to the organization with the property org:organization. The modeling approach is similar to the W3C's PROV model, but using different property names. Bio CRM's property bioc:inheres in is used for representing both atemporal (unary roles and binary relationships without qualifiers) and temporal (qualified by using events) roles of people. This is an informed decision for the simplicity of the model. A different approach has been chosen in Basic Formal Ontology (BFO) 2.0, where relations can be represented as continuous or occurrent, with separate relation types for them (Smith et al., 2015). BFO's approach has been criticized for its complexity, that causes logic and usability issues (Mungall, 2013).
The development of Bio CRM was started in the EU COST project "Reassembling the Republic of Letters" 16 (Tuominen et al., 2018).