adap ve methods for user1centered organiza on of music
play

Adap%ve(Methods(for(User1Centered( Organiza%on(of(Music(Collec%ons( - PowerPoint PPT Presentation

Data & Knowledge Engineering Group Adap%ve(Methods(for(User1Centered( Organiza%on(of(Music(Collec%ons( Doctoral(Thesis(Defense((Sebas%an(Stober( Magdeburg(|(November(15,(2011( The(Vision( ! an(intelligent(soIware(to(help(me( !


  1. Data & Knowledge Engineering Group Adap%ve(Methods(for(User1Centered( Organiza%on(of(Music(Collec%ons( Doctoral(Thesis(Defense(–(Sebas%an(Stober( Magdeburg(|(November(15,(2011(

  2. The(Vision( ! an(intelligent(soIware(to(help(me( ! organize(my(music(collec%on( ((no(simple(structuring(by(meta1data(but(by(similarity)( ! find(music,(I(like(to(listen(to(in(a(specific(moment( ((no(query1result1lists(but(explora%on)( The(1 st (Problem( listening(example:( ! How(should(the(soIware(compare(these(songs?( ! melody,(mood,(%mbre,(lyrics,(tempo,(dynamics…( ! mode,(instrumenta%on,(key,(harmonics,(rhythm,(meter(…( ➟ music(has(many(facets(–(How(important(is(each(one?( ➟ How(can(the(soIware(learn(how(I(compare(songs?( Sebas%an(Stober(1(Adap%ve(Methods(for(User1Centered(Organiza%on(of(Music(Collec%ons(|(Doctoral(Thesis(Defense,(OvGU(Magdeburg( November(15,(2011( 2(

  3. The(Thesis( ! introduc%on(to(music(informa%on(retrieval((MIR)( ! state(of(the(art(in(adap%ve(MIR( ! fundamental(techniques( ! data1adap%ve(feature(extrac%on( ! user1adap%ve(genres( ! context1adap%ve(music(similarity( ! focus1adap%ve(visualiza%on( ! bisocia%ve(music(discovery( ! gaze1controlled(adap%ve(focus( ! conclusion(&(outlook( Sebas%an(Stober(1(Adap%ve(Methods(for(User1Centered(Organiza%on(of(Music(Collec%ons(|(Doctoral(Thesis(Defense,(OvGU(Magdeburg( November(15,(2011( 3(

  4. Music(Informa%on(Retrieval((MIR)( “the%interdisciplinary%science%% % % % %of%retrieving%informa5on%from%music”% [wikipedia]( music(data( user( “His(Master's(Voice”((Francis(Barraud)( Sebas%an(Stober(1(Adap%ve(Methods(for(User1Centered(Organiza%on(of(Music(Collec%ons(|(Doctoral(Thesis(Defense,(OvGU(Magdeburg( November(15,(2011( 4(

  5. Music(Informa%on(Retrieval((MIR)( “the%interdisciplinary%science%% % % % %of%retrieving%informa5on%from%music”% [wikipedia]( a(typical(MIR(system:( data(interface( core(retrieval(system( user(interface( retrieval(model( structuring ( (music(data( index(/(DB( querying ( similarity( (user( ranking ( feature( extrac%on ( presenta%on ( preferences( classifica%on ( Sebas%an(Stober(1(Adap%ve(Methods(for(User1Centered(Organiza%on(of(Music(Collec%ons(|(Doctoral(Thesis(Defense,(OvGU(Magdeburg( November(15,(2011( 5(

  6. MIR(Challenges( [Downie’03] ( music(is(mul%1cultural( music(informa%on(has(many(facets(( ((((((((((and(can(be(represented(in(mul%ple(ways(( & \ \ D D 1 . Q . 3 % \ E E . . Q . ! ! \ . . . . (((((users(of(MIR(systems(have(different(musical(backgrounds( and(have(varying(informa%on(needs(( music((can(be(experienced(in(many(ways(( (((((((((((leading(to(different(percep%ons( Sebas%an(Stober(1(Adap%ve(Methods(for(User1Centered(Organiza%on(of(Music(Collec%ons(|(Doctoral(Thesis(Defense,(OvGU(Magdeburg( November(15,(2011( 6(

  7. Adap%ve(Systems( A(system(is((context)( adap5ve% iff( 1) it(behaves(different(in(different(contexts(given(the(same( input( [based(on(Broy(et(al.(‘09] ( AND( 2) the(respec%ve(adapta%on((i.e.,(the(difference(in(behavior)( is(goal1driven(in(that(it(aims(to(op%mize(the(system’s( behavior(in(the(given(context(according(to(some(pre1 defined(measure.( Sebas%an(Stober(1(Adap%ve(Methods(for(User1Centered(Organiza%on(of(Music(Collec%ons(|(Doctoral(Thesis(Defense,(OvGU(Magdeburg( November(15,(2011( 7(

  8. Adaptable( ➞ (Adap%ve(System( INPUT OUTPUT adaptable system control parameters adaptation logic evaluate USER context model adaptive system environment context sensing Sebas%an(Stober(1(Adap%ve(Methods(for(User1Centered(Organiza%on(of(Music(Collec%ons(|(Doctoral(Thesis(Defense,(OvGU(Magdeburg( November(15,(2011( 8(

  9. A DAPTIVE (M USIC (S IMILARITY ( data interface core retrieval system user interface retrieval model structuring music data index / DB querying similarity user ranking feature extraction presentation preferences classification Sebas%an(Stober(1(Adap%ve(Methods(for(User1Centered(Organiza%on(of(Music(Collec%ons(|(Doctoral(Thesis(Defense,(OvGU(Magdeburg( November(15,(2011( 9(

  10. Goal(/(Problem(Formula%on( Learn(mul%1facet(music(similarity(measures(( that(reflect(the(user’s(informa%on(need(and(context!( Sebas%an(Stober(1(Adap%ve(Methods(for(User1Centered(Organiza%on(of(Music(Collec%ons(|(Doctoral(Thesis(Defense,(OvGU(Magdeburg( November(15,(2011( 10(

  11. Adaptable(Model(of(Similarity( ! objects(of(interest(are(described(by(various(features(( ! capture(different(aspects(of(similarity( ! may(not(be(equally(important(for(comparison( ! distance(facet(( =((set(of)(feature(s)(( " objec%ve( +(distance(measure( ! non1nega%ve: (d(a,b)(≥(0([and(d(a,b)(=(0(iff(a=b]( ! symmetric: (d(a,b)(=(d(b,a)( ! op%onally:( (fulfills(triangle(inequality( ➟ distance(=(weighted(linear(sum(of(facet(distances( ! weights(non1nega%ve,(constant(weight(sum( ! direct((manual)(adapta%on(possible( " subjec%ve( (simple(&(understandable(model)( Sebas%an(Stober(1(Adap%ve(Methods(for(User1Centered(Organiza%on(of(Music(Collec%ons(|(Doctoral(Thesis(Defense,(OvGU(Magdeburg( November(15,(2011( 11(

  12. System(Design( objec%ve( subjec%ve( INPUT ( aggregated(distance( OUTPUT ( facet(distances( adaptable(system( distance(measure( control(parameters( control(facet(weights( adapta%on(logic( evaluate( USER ( context(model( rel. ( distance(constraints( user(ac%ons(/( adap%ve(system( expert(annota%ons( environment( context(sensing( Sebas%an(Stober(1(Adap%ve(Methods(for(User1Centered(Organiza%on(of(Music(Collec%ons(|(Doctoral(Thesis(Defense,(OvGU(Magdeburg( November(15,(2011( 12(

  13. General(Adapta%on(Approach( user(ac%ons( derive( rela%ve(distance(constraints( as(constraints( as(training(examples( op%miza%on(problem:( classifica%on(problem:( • find((valid)(weights(that( • learn(linear(classifier(for(( a) sa%sfy(all(constraints( +(training(examples( d(s,a)%<%d(s,b)% b) minimize(error((#viola%ons)( – (training(examples( d(s,a)%>%d(s,b)% • weights(are(defined(by(the(separa%ng( hyperplane( [Cheng(&(Hüllermeier(‘08] ( Sebas%an(Stober(1(Adap%ve(Methods(for(User1Centered(Organiza%on(of(Music(Collec%ons(|(Doctoral(Thesis(Defense,(OvGU(Magdeburg( November(15,(2011( 13(

  14. Facet(Weight(Adapta%on(Approaches( ! Gradient(Descent((op%miza%on)( ! directly(minimizes(error((constraint(viola%ons)( ! problem:(may(get(stuck(in(local(minimum( ! Quadra%c(Programming((op%miza%on)( ! minimizes(weight(change(subject(to(( ! hard(weight(bounds(and( ! hard(or(soI(distance(constraints((addi%onal(slack(variables)( ! con%nuity((no(abrupt(changes)( ! Linear(Support(Vector(Machine((classifica%on)( ! maximizes(margin((between(+(and(–(training(examples)( ! favors(“stable”(solu%ons(( ! problem:(soI(weight(constraints(may(be(violated((neg.(weights)( Sebas%an(Stober(1(Adap%ve(Methods(for(User1Centered(Organiza%on(of(Music(Collec%ons(|(Doctoral(Thesis(Defense,(OvGU(Magdeburg( November(15,(2011( 14(

  15. Applica%ons(&(Considered(User(Ac%ons( ! Liederenbank( [ISMIR’09] ( ! classifying(Dutch(folk(songs( ➟ class(annota%ons((by(experts)( [hvp://www.liederenbank.nl]( ! BeatlesExplorer( [AMR’08] ( ! structuring(the(Beatles(dataset( ➟ moving(songs(to(other(cells( ➟ correc%ng(similarity(rankings( ! MusicGalaxy( [CMMR/SMC’10] ( ! exploring(media(collec%ons( ➟ tagging(objects( Sebas%an(Stober(1(Adap%ve(Methods(for(User1Centered(Organiza%on(of(Music(Collec%ons(|(Doctoral(Thesis(Defense,(OvGU(Magdeburg( November(15,(2011( 15(

  16. F OCUS DA DAPTIVE (V ISUALIZATION ( data interface core retrieval system user interface retrieval model structuring music data index / DB querying similarity user ranking feature extraction presentation preferences classification Sebas%an(Stober(1(Adap%ve(Methods(for(User1Centered(Organiza%on(of(Music(Collec%ons(|(Doctoral(Thesis(Defense,(OvGU(Magdeburg( November(15,(2011( 16(

  17. Mo%va%on( ! generate(an(overview(of(a(music(collec%on(for(explora%on( ! idea:(use(dimensionality(reduc%on(techniques( high1dimensional( 2D(display( feature(space( MusicMiner( SoundBite(for(Songbird( Islands(of(Music( ([Mörchen(et(al.(2005]( ([Lloyd(2009]( [Pampalk(et(al.(2003]( Sebas%an(Stober(1(Adap%ve(Methods(for(User1Centered(Organiza%on(of(Music(Collec%ons(|(Doctoral(Thesis(Defense,(OvGU(Magdeburg( November(15,(2011( 17(

  18. Focus1Adap%ve(SpringLens( temporarily(fix(/(highlight(( the(neighborhood(in(focus( Sebas%an(Stober(1(Adap%ve(Methods(for(User1Centered(Organiza%on(of(Music(Collec%ons(|(Doctoral(Thesis(Defense,(OvGU(Magdeburg( November(15,(2011( 18(

  19. Focus1Adap%ve(SpringLens*( ! mul%1focus(fish1eye(distor%on(highlights(nearest(neighbors( ! primary(lens( ! controlled(by(user( ! enlarges(region(of(interest( ! more(space(for(details( ! preserves(context( ! secondary(lenses( ! data1driven( ! highlight(nearest(neighbors(( ! show(“wormholes”( ! neighbors(come(closer( *based(on(SpringLens(non1linear(distor%on(technique([Germer(et(al.(‘06]( Sebas%an(Stober(1(Adap%ve(Methods(for(User1Centered(Organiza%on(of(Music(Collec%ons(|(Doctoral(Thesis(Defense,(OvGU(Magdeburg( November(15,(2011( 19(

  20. System(Design( dataset(projec%on( INPUT ( distorted(projec%on( OUTPUT ( adaptable(system( SpringLens( posi%ons( control(parameters( control(lens(parameters( neighborhoods( adapta%on(logic( evaluate( USER ( context(model( focus(model( region(of(interest( adap%ve(system( (primary(lens)( environment( context(sensing( Sebas%an(Stober(1(Adap%ve(Methods(for(User1Centered(Organiza%on(of(Music(Collec%ons(|(Doctoral(Thesis(Defense,(OvGU(Magdeburg( November(15,(2011( 20(

Recommend


More recommend