This paper introduces DOREMUS|a semantic web project aiming to provide common knowledge models and shared multilingual vocabularies to cultural institutions, publishers, distributors and users in the musical domain. The project develops methods to describe, publish, connect and contextualize music catalogs on the web of data. Our focus is on the description of classical and traditional musical works as well as their interpretations (events).
Music is everywhere|played, recorded, broadcasted and listened to. Files
of recorded music are all over the web|stored, streamed, shared or sold. The
knowledge about musical works is described in detail in the information systems
of several cultural and media organizations around the world. The leading actors
in France|the BnF (Bibliotheque Nationale de France), the Philharmonie de
Paris and Radio France, as well as the Meaning Engines company join forces
with three academic institutions within the DOREMUS project1 in order to
make this knowledge available and re-usable on the web of data. The project
covers three main objectives: (
In this paper, we describe our approach for publishing and interlinking
descriptions of musical data based on ontological models and vocabularies and we
discuss several applications. The goal is to exploit as much as possible the rich
data which is available and to open it to the public. The main contributions
consist in (
Musical works are complex objects. Expressing them comprehensively requires the description of their physical manifestations (recordings, scores) and all the events that de ne them (creation, publication, performance). The rst aspect is relatively well-mastered, in library catalogues as well as in the music industry. Several models can be used to describe a musical work, some being speci c to the music domain (MusicOntology), while others being more broadly designed for libraries in general (FRBR2, UNIMARC3). The second aspect is rather new, although there is a growing need and interest in using even-based models. Several ontologies aim speci cally to de ne events (Event4, LODE5), but there are fewer examples of using them to describe the creation and publication processes.
One of the di culties with musical works is that although their expressions may di er signi cantly, they are still regarded as a single work. Modeling it requires to express the singleness of the work as well as the speci cities of its expressions, and to show how events are connected to these expressions. Another issue is that an arrangement may be considered as an expression or as a new work, depending on the data producer. Our model shall also make it possible to specify the relations between works, for instance when one work is derived from another.
We are currently working on a model enabling the expression of all these aspects in a coherent way, while being exible and powerful enough to interoperate 2 http://www.ifla.org/publications/functional-requirements-for-bibliographic-records 3 http://www.ifla.org/publications/unimarc-formats-and-related-documentation 4 http://purl.org/NET/c4dm/event.owl 5 http://linkedevents.org/ontology/ with information systems dealing with any kind of cultural data. It occurred to us that the FRBRoo ontology had all the required qualities. It is based on the FRBR and CIDOC-CRM models, with a concern for the necessity to describe how events occur through the life of a thing and its resulting states. It is a formal ontology, based on the articulation of events and things, containing the notion of work, and o ering precise elements to describe physical items. Its structure is adaptable and allows to model cultural metadata with ne granularity. Since the work, as de ned in this model, is not speci cally musical, we are extending the model by additional classes and properties that are recurrent and important in the description of musical works (Figure 1). 3
As in many other domains, the DOREMUS project faces the problem of processing a very large amount of heterogeneous (in terms of formats and languages) and semi-structured data in order to transform them into RDF, describe them by (re-)using appropriate vocabularies, link these data to existing datasets and publish them on the web of data|a process also coined as data lifting [1]. We have followed and adapted the work ow of the Datalift platform (Figure 2). The methods for extracting RDF triples need to be adapted to the data source in the musical eld, which consists of documents in the MARC (MAchine-Readable Cataloging) format, as well as XML documents.
There exist several approaches for extracting RDF triples from XML documents [2]. However, MARC (and its variants) documents has, so far, been rarely processed by semantic web applications. We have developed a two steps approach for transforming MARC data into RDF, while dealing with two variants of the MARC format: UNIMARC and INTERMARC. First, we propose an approach to extract raw RDF at the highest possible level of granularity by exploiting subareas coded in MARC with an alphanumeric character and a delimiter "$" (Figure 3). The principle is that each area and sub-area will be represented by an RDF property and the RDF object has the value of the area or one of its subelds. Second, we perform a manual mapping between the UNI/INTER-MARC areas and sub-areas with classes and properties de ned in the models discussed in Figure 1. Thus, the description of musical works is performed at a lower level of granularity (due to merging of certain sub elds taken as values of a property), while ensuring completeness of the information.
Several documents within a single institution or across institutions describe the same entity (a particular musical work), but from di erent perspectives (scores, sound recordings, videos). We aim to develop generic data connectors [1] to automatically detect relationships (equivalence, transcription, interpretation) between musical entities. The multilingual nature of data|a characteristics that should be preserved in the data lifting process|will be handled by automating the alignment process of multilingual ontologies speci c to the eld of music. 4
Once the musical data coming from the cultural institutions is on the web of data, we plan to develop a musical recommendation application. It will serve as an evaluation of our model, and enable us to control the quality of data, after being modeled, lifted and interlinked. It will also be an example of how to implement the DOREMUS model for other cultural institutions.
The application will o er features such as the creation of original playlists for specialized radio, or the selection of appropriate musical works and interpretations to illustrate the biography of a composer, a historical period, culture or genre. The targeted end-users range from a child building his/her musical culture, through music lovers or curious discoverers, to expert musicologists.
This work has been partially supported by the French National Research Agency (ANR) within the DOREMUS Project, under grant number ANR-14CE24-0020.