The Networked Event Model
Over 2012-3 the SSHRC Funded project Humanities RDF explored how humanities researchers might employ the Research Description Framework (RDF) and Named Entity Extraction (NEE) to document the connections between actors, objects, and interactions evident in the prose work of a large interdisciplinary research project, Making Publics. It focused on the degree to which extracted people, places, and things, could be linked and presented in a format that would to allow researchers to effectively disseminate their work in the world of Linked Open Data. This work resulted in the Networked Event Model (NEM), a prototype for describing forms of social interaction between agents (widely construed) and objects or people. In its simplicity, NEM sees all interactions as the heart of socio-cultural processes, viewing ‘events’ as the core data required for documenting complex social network models that can account for change over time, as well as be flexible enough to handle contestation and issues of provenance. Humanities scholars involved in the development of NEM saw the inherent linguistic utility of RDF, but were stumped by the need of many event models to reify or instantiate social events as if they were clearly defined objects or nouns, rather than processes bounded by starts and stops, and how agency might be preserved within a social networking data model. The use of quads radically alters the RDF modelling approach, as it turns all relationships between entities into events which can be clearly identified and documented as assertions, thus resolving a key problem in RDF – that of provenance. But the utility extends much further as NEM’s hermeneutic takes up the triple’s syntactical structure as a means for documenting social agency in RDF: by focusing on verbs and processes of interaction, NEM allows the documentation of interactions evident in prose in the form of data. Using identifiers not only solves the provenance issue, it also permits the nesting of events, turning events themselves into entities that can be referenced.
NEM is innovative in how it harnesses the inherent structure of the RDF triple as a syntactical assertion that can also capture agency on the one hand, and on the other in how an addition of an identifier allows the concatenation of events into more complex assertions about social interaction. In short, NEM allows clear documentation of interactions by mimicking linguistic structures to a certain degree, permitting scholars to document interconnections within historical evidence in a way that reflects their social complexities. At the core of the prototype model lies the event entity, comprising a simple named graph with the following composition: ID: Agent->verb->object. NEM limits agents and objects to people, organizations, places, and things and provides a restrictive list of verbs. As its primary purpose is to provide a means of documenting concrete forms of interaction, this restrictive vocabulary focuses on clear assertive actions like ‘wrote’, ‘killed’, ‘sued’ rather than the subjective, affective, or epistemological such as ‘knows’, ‘might have eaten’, ‘loves’ etc. Beyond the event entity, NEM also uses well-established vocabularies like Dublin Core, CIDOC-CRM, and Geonames to represent its other entities. It is not a top-level ontology, but a mediating schema specifically engineered to map interactions and associations between biographical, archival, and geospatial data over time, seeing such relationships – events – as mutable and limited processes which arise and dissipate.
An important element of the project was experimental event recognition using NEE and the General Architecture for Text Engineering (GATE). The basic event model recognition pattern was:
<EdgeSentence> <Person1></Person1> <Edge> <Verb></Verb> </Edge> <Person2></Person2> </EdgeSentence>
The current verb list is available through the Nanohistory website at http://www.nanohistory.org/about/specs/verbs/