Representing musicology knowledge on the Web using Linked Data

Transcription

1 Representing musicology knowledge on the Web using Linked Data The Semantic Web and Why You Should Care Digital Humanities at Oxford Summer School Kevin Page Oxford e-research Centre, University of Oxford.

2 Good morning! This session: An overview of the the Semantic Web, from the perspective of a scenario in music. An introduction to a system built to serve that scenario Illustrating both the general ideas, and the technical underpinnings. More Linked Data to follow later in the week...

3 Motivation: in general When knowledge has been generated, we should capitalise on its value by capturing it publishing it using it linking it re-using and building upon it ( unintentionally?)

4 A digital musicologist might Building upon previous method and output data... Take an input data set Develop, combine, or finesse a method or process Produce findings, results and output data...which can be used, combined (and improved?)

5 It's about making your output useful to others......and building upon the output of others to make your own work better

6 ...or, at least, easier 6

7 On the Web...? (technically speaking!) Don't just put Digital Musicology content on the Web......but use and build upon Web Architecture to scale Digital Musicology activity The value is in the linking.

8 A brief history of the Semantic Web 8

9 Vanevar Bush and the memex 9

10 Doug Engelbart He envisioned intellectual workers sitting at display 'working stations', flying through information space, harnessing their collective intellectual capacity to solve important problems together in much more powerful ways. Harnessing collective intellect, facilitated by interactive computers, became his life's mission at a time when computers were viewed as number crunching tools. 10

11 Ted Nelson and hypertext 11

12 Tim Berners-Lee and the World Wide Web This machine is a server DO NOT POWER IT DOWN!! 12

13 So what, then, is the Semantic Web? The web is the largest and most successful distributed system ever constructed But it is a Web of Documents The Semantic Web is the effort to create the equivalent Web of Data

14 Semantic Web activities have been baking since the late 1990s The infamous layer cake has evolved Linked Data is a more recent movement...

15 Linked Data There are two words in Semantic Web Both are important!

16 So how's it going?

17 SW vs. LD vs. LOD vs. OD The World Wide Web Consortium (W3C) Semantic Web activities and standards define the underlying technologies. Linked Data is a more recent movement. Data can be Open, without being Linked. Data can be Linked, without being Open. 17

18 Linking Up

19 Musicbrainz & Linkedbrainz Aligning Musicbrainz NGS with Linked Data Centre for Digital Music at Queen Mary University of London

20 Musicnet Canonical URIs for composers Alignment tool for managing co-references University of Southampton

21 Sonic Visualiser & Annotator Uses RDF for serialisation Numerous associated ontologies for tasks Centre for Digital Music at Queen Mary University of London

22 BBC Music Pages generated using, and available as, RDF Linked to other data sources

23 dbtune.org/* Several music related datasets converted and published as Linked Data Linked to other data sources SPARQL endpoints

24 Digital Music Lab

25 ...and more to come SALAMI (Wednesday) Early Music (Friday) Wagner (Friday) In Concert (Friday) Not necessarily Linked Data projects but musicology making use of Linked Data Digital Musicology is a fantastic Linked Data opportunity!

26 The basics Subject Predicate Object That's it. An RDF triple (RDF is the Resource Description Framework) 26

27 Triples of information, linked to the scale of the Web lived in Wolfgang Amadeus Mozart born in lies within Land Salzburg (State) occupation composed The Marriage of Figaro Salzburg lived in a land of Composer performed in premiered in Vienna capital of Austria Is a Opera 27

28 Linked Data considerations The previous example was very simplistic The combination of many specialised fields of scholarship will form a large and complex web Linked Data is about making associations between information with common meaning (semantics) Not (necessarily) about forcing common meanings across information Linked Data is a framework for scalable semantic interoperability of digital information. It can aid scholarly investigation, but doesn t remove the need for it! 28

29 Simultaneous contexts RDF, as a standard, can accommodate multiple simultaneous contexts (perspectives, specialisms) which are formed through the practice of a community Nurmikko-Fuller, T., Page, K. et al. Building Complex Research Collections in Digital Libraries: A Survey of Ontology Implications ACM/IEEE Joint Conference on Digital Libraries (2015) 29

30 Linked Data examples So the basic principles and technologies are very simple But a large number of specialisms and data sources each providing (very simple) Linked Data combine to form a huge and (overall) complex web of data The power of the technology and tooling is in allowing us to cope with that complexity Though this proves tricky to teach by example: so we will work through a simplified but real Linked Data system... 30

31 A simple example

32 A simple example

33 A digital musicologist might Building upon previous method and output data... Take an input data set Develop, combine, or finesse a method or process Produce findings, results and output data...which can be used, combined (and improved?)

34 The How Country is my country? example Use artist and location metadata to select a collection(s) of audio Perform a genre classification over the collection Publish the genre analysis data with links back to the tracks (and so artist, location, and collections) Combine the results with other published metadata (about the artist, location, collections)

35 A simple example (simplified!) Use artist and location metadata to select a collection of audio 1 Location 5 Artist 2 (Digital) Recordings Performance Musical Work 3 Perform a genre classification over the collection 4 Genre analysis...and publish the analysis results with links back to the metadata Composer Lyrics Music analysis Structural analysis

36 Video of the example application A digital musicologist using the Country/country system

37 Ttable Why are a music artist Depression is Ridiculous is an Album Passive Defiance ishas a track trackon Depression is Ridiculous Ttable Why made Depression is Ridiculous Warning: approximated for clarity!

38 RDF Triples joined together form a graph Ireland are from Ttable Why are a music artist is an Album made 6385.mp3 available Depression available as is Ridiculous is a track on Passive Defiance Warning: approximated for clarity!

39 URIs everywhere! Very significantly, each element of the triple can be a URI a Uniform Resource Identifier e.g. Ttable Why are a music artist

40 Benefits of URIs This means we are always clear whether we are referring to the same Thing, or property, or concept,... The graph can be split up and distributed while retaining consistency (if desired) We can also use HTTP URIs as a mechanism for retrieving parts of a graph, It's a web of data!

41 How do we express our RDF? 41

42 RDF is not RDF/XML RDF is an abstract model with several serializations. RDF/XML is one of them, but... it's ugly. RDF/XML is for our friends the machines, not for us. Other serialisations include RDFa, N-Triples, JSON-LD, and... 42

43 Turtle Much more suitable for humans hacking RDF because it is awesome [1] 43 [1]

44 @prefix rdf: mo: rdfs: dc: ns1: < < mo:track < < mo:published_as < < rdf:type mo:track, rdfs:resource ; mo:available_as < < ; dc:title "Disenchantment Anguish"^^< ; mo:license < ; mo:track_number "3"^^<

45 Where do we put our RDF? Where do we get RDF from? 47

46 GET it GET and request an RDF content type, and rdf: mo: rdfs: dc: ns1: < < mo:track < < mo:published_as < < rdf:type mo:track, rdfs:resource ; mo:available_as < 48

47 RDF In a triplestore RDF can also be stored in a triplestore This can also be thought of as a cache where triples are brought together to do useful things to them as a whole e.g. reasoning over the triplestore

48 Data & service distribution Wikipedia BBC Geonames DBTune Location Artist Musicbrainz Jamendo (Digital) Recordings Performance Linkedmusic.org (us!) Genre analysis Musical Work Composer Lyrics Music analysis Structural analysis

49 Genre Classifier Output e.g. We took the Downie lab's Son of Blinkie classifier which runs in the Meandre workflow framework Used myexperiment to run the workflow over all signal in a collection as created by the Collection Builder Modified the output to serialise as RDF with genres aligned (where possible) with dbpedia RDF output is published and held in a triplestore

50 Genre Classifier Output: RDF jamendo/signal/ mo:available_as audiofiles.linkedmusic. org/audiofile/98933 sim:element sim:weight _:sim1 a sim:similarity sim:element sim:method resource/country_music sim:workflow GenreAnalysis Warning: approximated for clarity!

51 RDF In a triplestore RDF can also be stored in a triplestore This can also be thought of as a cache where triples are brought together to do useful things to them as a whole e.g. reasoning over the triplestore Most usefully from our perspective, we can query such collections of RDF through a powerful interface called SPARQL 56

52 SPARQL The SPARQL Protocol and RDF Query Language SPARQL has several queries, the main one being SELECT Let's just stick to the query language Also CONSTRUCT, ASK, and DESCRIBE As with Turtle, we start a query by defining prefixes 57

53 Thinking of SPARQL queries at triple patterns? composed?* premiered in performed in Find anythings performed in Vienna Find operas performed in Vienna Find operas premiered in Vienna Find operas premiered in capitals of Austria Find composers of operas premiered in capitals of Austria * Vienna capital of Austria is a Opera 60

54 Triples of information, linked to the scale of the Web lived in Wolfgang Amadeus Mozart born in lies within Land Salzburg (State) occupation composed The Marriage of Figaro Salzburg lived in a land of Composer performed in premiered in Vienna capital of Austria Is a Opera 61

55 Triples of information, linked to the scale of the Web lived in?= born in Wolfgang Amadeus Mozart (The Marriage of Figaro) lies within Land Salzburg (State) occupation composed * Salzburg lived in a land of Composer performed in premiered in Is a * (Vienna) capital of Austria Opera 62

56 In our example: Collection Builder To begin with, we create our collections using metadata only We use SPARQL to query an endpoint for tracks by artists from a particular country We publish our collection as RDF We can then query our Audio File Repository to find any locally available signal in the collection, and publish this as a derivative collection 63

57 SPARQL Queries PREFIX foaf: < PREFIX mo: < PREFIX dc: < SELECT?artistname WHERE {?artist a mo:musicartist ; foaf:name?artistname ; foaf:made?album.?album a mo:record ; dc:title "Depression Is Ridiculous". } 64

58 Collection Builder: SPARQL 65

59 How do we exchange meaning through our RDF? Ontologies. 66

60 What is an ontology? An ontology is a description of concepts and their relationships It enables us to build semantic models with RDF (more specific models than RDF itself) It's about adding meaning to your data so that it can be understood and reused 67

61 So you just need to... Get the model right... Get a model right... Get your model right... 68

62 Avoiding rabbit holes 69

63 Appropriate Ontologies Ontologies don't remove complexity, but they do enable us to scale it (relatively) gracefully. More than one correct ontology can be applicable to a resource it depends what you're doing with it. Where available, use an applicable existing ontology (or extend it). Write the ontologies you need you can extend them later. Better it be limited and right... It is probably unwise to expect an ontology for all Things... 70

64 What you might want an ontology for Your input data. Your output data. Your method. 71

65 Ontologies you've already encountered in our example... The Music Ontology Friend-of-a-Friend (FOAF) Dublin Core dbpedia 72

66 Ontologies in our simple example Geonames Location Artist Music Ontology Musicnet (Digital) Recordings Performance OAI Object Reuse and Exchange Musical Work Lyrics Genre analysis Music analysis Segment Ontology Structural analysis dbpedia Composer Research Object (workflow)

67 How do we express our ontologies? (why, in RDF of course!) 74

68 RDFS and OWL RDFS: RDF Schema The basics required to structure an ontology and exchange vocabularies Classes and properties, super- and subclasses, range and domain OWL: Web Ontology Language More sophisticated structures Constraints for existence and cardinality, transitive, inverse, symmetrical properties,... 75

69 Ontology Examples from the Music Ontology (MO) MO extends and builds upon other ontologies: Event (and hence Timeline), FOAF, Dublin Core,... The Music Ontology, Yves Raimond et al, 76

70 Ontology examples from MO mo:analogsignal rdf:type owl:class ; rdfs:isdefinedby mo: ; rdfs:label "analogue signal" ; rdfs:subclassof mo:signal ; owl:disjointwith mo:digitalsignal. mo:recording rdf:type owl:class ; rdfs:isdefinedby mo: ; rdfs:subclassof event:event. Warning! Curtailed for illustrative purposes. 77

71 In our example: Results Viewer The output from our genre classifier(s), the Collection Builder(s), and Audio File Repository(ies) share a common underlying data model, identifiers, and vocabularies The web of links between them allows us to combine their information in a Results Viewer Links to external data sources and use of established ontologies allows us to easily incorporate other data in the Results Viewer 79

72 Results Viewer e.g. find all genre analysis of tracks in a collection, where the collection is of artists from a specific country Provide and view examples of other artists in this genre For the most heavily weighted track in this genre find audio files for playback 80

73 Results Viewer 81

74 Results Viewer SPARQL PREFIX dbpedia-owl: PREFIX foaf: SELECT?artist?artistname?place?placename WHERE {?artist dbpedia-owl:genre <$genreuri>. {?artist a dbpedia-owl:band ; dbpedia-owl:hometown?place. } UNION {?artist a dbpedia-owl:musicalartist ; dbpedia-owl:birthplace?place. }?artist foaf:name?artistname.?place foaf:name?placename. } Endpoint: 82

75 Summary We can use Linked Data to publish the output of Digital Musicology studies This enables us to scale and link different types of work We can use RDF and ontologies to capture, scale, and link domain knowledge A good Semantic Web application is just a better Web application!