Undefined 0 (2012) 1–0 1 IOS Press Hello Cleveland! Linked Data Publication of Live Music Archives Sean Bechhofer a;∗ David De Roure b Kevin Page b a School of Computer Science, The University of Manchester, United Kingdom, [email protected] b Oxford e-Research Centre, University of Oxford, United Kingdom fkevin.page,[email protected] Abstract. We describe the publication of a linked data set exposing metadata from the Internet Archive Live Music Archive. The collection provides access to recorded performances and is linked to existing musical and geographical resources. The dataset contains over 17,000,000 triples describing 100,000 performances by 4,000 artists. Keywords: Linked Dataset, Music, Internet Archive 1. Introduction more, in live situations artists will frequently play works by other artists (“covers”), providing source The Internet Archive Live Music Archive [12] (fur- content for cover detection algorithms [11]. ther referred to here as LMA) is an online resource Collection generation and management is a key providing access to a large community-contributed starting point for research in the digital humanities. An collection of live recordings. Covering nearly 4,000 earlier prototype [15] demonstrated how Linked Data artists, chiefly in rock genres, the archive contains over can be applied to the MIR research process and the 100,000 live recordings made openly available with utility of this approach, particularly when gathering the permission of the artists concerned. Audio files and managing corpora of source audio; however, this are available in a variety of formats, and each record- system re-used pre-existing Linked Data that described ing is accompanied by metadata describing informa- the recordings to populate its collections. As compu- tion about dates, venues, set lists, the provenance of tational analysis increases in scale through projects the audio files and so on. such as SALAMI [4], so too does the value of re- From a musicological perspective, the collection is publishing existing large repositories such as the LMA valuable for a number of reasons. First of all, it pro- using Linked Data: as it stands, however, extracting vides access to the underlying audio files. Thus the subcollections from the archive is not a straightforward LMA provides a corpus that can be used for Music task. Metadata is largely published as free text fields, Information Retrieval (MIR) [3] tasks such as genre with heterogeneity in detail and inconsistency in con- detection, key detection, segmentation and so on as tent. Providing structured metadata (with links to ex- exemplified by the MIREX series of workshops [5]. ternal resources) will, we believe, facilitate the activi- It provides multiple recordings by individual artists1 allowing comparisons across performances. Further- ties such as the production of sub-corpora for experi- ments or evaluation. We describe an exercise in republishing LMA meta- * Corresponding author data using a linked data [2] approach. We believe 1In the case of the Grateful Dead, an act that for many years en- couraged audience taping of performances, the LMA contains over that publishing the collection metadata as linked data 8,000 recorded performances. brings benefits in supporting query and integration 0000-0000/12/$00.00 c 2012 – IOS Press and the authors. All rights reserved 2 with existing sources. Note that we are dealing here driven curation and validation. Explicit separation of purely with the metadata, and leave the audio source the “ground facts” from the alignments also allows for files untouched (but provide links to the online re- revisions and additions to those alignments. sources). Enhancing the collection with additional in- formation gained through analysis of those audio files 3. Modelling is left to future work. The collection contains a number of basic entities. 2. Approach Artist The performer of a concert/gig. There is no at- The collection is published using a layered ap- tempt to link or split up bands, duos, etc. proach. The core metadata describing the resources is Performance A particular concert/gig that has been essentially published “as is”. Raw data provided by recorded. Each performance is given a unique LMA is translated to an RDF form, using appropri- identifier by LMA. ate vocabulary terms (for example, the label associ- Track Performance of a particular track within a Per- ated with a particular performance is represented us- formance ing skos:prefLabel). Additional information as- Venue Location where a performance takes place. serting mapping relationships to other collections such as MusicBrainz [14], GeoNames [6] or last.fm [9] is Each Artist, Performance, Track and Venue is minted then added. For example, although attempts could have a URI in the collection namespace2 with an appropri- been made to reconcile artist names as used in the ate path prefix. A number of ontologies are used for collection, this is not achieved through modification the description of entities. of the core data. This method allows us to explicitly record provenance information about how the associ- Music The Music Ontology [16,18] provides terms ations were derived, which in turn then allows con- that describe performances, artists and the rela- sumers of the data to make decisions about whether or tionships between them. not to use or trust the relationships asserted. It is thus Events The Event Ontology [17] provides terms for clear to any consumer of the data whether information describing events. has come directly from LMA or is additional informa- Similarity The Similarity Ontology [7,8] provides tion provided via our process. We believe that such an terms for asserting associations between entities. approach is needed for a collection like this, where the This is used to associate artists in the collection data is not simply “asserted truth”, but has some sub- with MusicBrainz ids, and locations with last.fm jectivity. venues and GeoNames entities. Note also that our approach explicitly avoids the use SKOS SKOS [13] labelling properties are used to la- of owl:sameAs triples to relate LMA entities to ex- bel entities. ternal entities (e.g. within MusicBrainz or GeoNames) PROV-O The W3C provenance ontology [10]. and instead uses a pattern from the Similarity Ontol- VoID The W3C dataset metadata ontology [1]. ogy (see Section 4). This is useful in supporting pro- etree An ontology3 that defines subclasses of Music cesses such as artist reconciliation, where matching be- Ontology classes and specific properties used in tween recording artists may be a somewhat fuzzy pro- the etree metadata. cess – groups or bands can be volatile organisations in terms of membership (although they do not appear The basic modelling pattern used within the data in the LMA, The Fall’s “revolving door of musicians” set is shown in Figure 1. In the figures, green, un- is a case in point [19]) and we may not always have labelled links are rdf:type. Blue, unlabelled links confidence that descriptions of an artist from differ- are rdfs:subClassOf. The ontology used to de- ent sources match exactly. Data consumers then have scribe the collection is relatively inexpressive, essen- the option of using the encoded raw source data or the tially providing classes for performances and venues additional layer of mapped relationships. Furthermore, and properties for the assertion of values and relation- layering allows for the possibility of further analysis ships. and re-publication of mapping relationships from the source corpus as new techniques and corrections are 2http://etree.linkedmusic.org developed; this could include publication community 3http://etree.linkedmusic.org/vocab 3 Fig. 1. Basic Model 4. Record Linkage further identification of groups or solo artists and a re- The collection offers possibilities for record linkage finement of the types applied (for example asserting with external datasets. In particular, music artists and that a resource is in fact a mo:MusicGroup). geographical locations are entities that are described in Track Alignment In the current dataset, tracks are not a number of external data sources (many of which are aligned to external data sources. See Section 9 for fur- also published as Linked Data). ther discussion. Artist Alignment MusicBrainz [14] provides an “open Geographical Alignment Performances occur at a music encyclopaedia” and provides identifiers for a particular place4 and can thus potentially be mapped large number of music artists. MusicBrainz is a clear to geographical locations in collection such as GeoN- candidate for linking from a collection like LMA. ames. Concert performances also tend to take place in Queries to MusicBrainz taking exact matches on specific venues (theatres, concert halls etc) which are names provides a simple alignment between our artists described in data sources such as last.fm [9]. Informa- and MusicBrainz, covering 1,168 of the 3,981 artists tion about venues and general locations is given in the in the collection. In keeping with the strategy outlined source metadata, with variable granularity and consis- in Section 2, the relationships between the artists and tency, using the venue and coverage tags: MusicBrainz are asserted using the Similarity Ontol- ogy. Figure 2 shows how these relationships are as- venue The name of the venue, e.g. concert hall, club, serted. festival. where the performance was recorded. The Music Ontology considers mo:MusicArtist coverage The larger geographical area for the loca- to encapsulate ”A person or a group of people [...], tion, e.g. city or state. whose musical creative work shows sensitivity and imagination” and the current dataset makes no distinc- The raw location information suffers from incon- tion between solo artists and bands/groups of musi- sistencies in presentation (e.g. Chicago, IL; Chicago, cians. There is no information in the source corpus that Il; Chicago, Illinois; Chicago etc.). Location informa- distinguishes between solo artist and group or identi- fies relationships between, for example, a singer and 4To the best of our knowledge, the collection does not contain a band.