Transcript Classifiers - McGill Schulich Faculty of Music

Usability
Fujinaga 2003
Uitdenbogerd, A., and R. Schyndel. 2002. A review of factors affecting music recommender
success. International Symposium on Music Information Retrieval. 204-8.
•
•
•
Design criteria for music recommender systems
Survey of research into musical taste
Review of music recommenders
– Provide personalized content to users
• Messages
• List of stories
• Artwork
– Collaborative filtering (collect users’ opinions, ranking)
– Content-based filtering
•
Limitations:
– Inadequate raw data (editorial information)
– Lack of quality control (user preference)
– Lack of user preferences for new recordings
• Content-based analysis needed for new recordings
– Presentation (mostly simple lists)
Uitdenbogerd, A., and R. Schyndel. 2002. A review of factors affecting music recommender
success. International Symposium on Music Information Retrieval. 204-8.
• Goals
– Simple to use with minimum of input
– More effort in providing input lead to better
recommendations
– Choice of music based on preferences, style, or mood
• Use existing research into factors affecting musical
taste
– Social psychology
– Demographics for marketing
Uitdenbogerd, A., and R. Schyndel. 2002. A review of factors affecting music recommender
success. International Symposium on Music Information Retrieval. 204-8.
•
Existing research
–
–
–
–
–
–
–
–
Stable extraverts: solid predictable music
Stable introverts: classical and baroque styles
Unstable extraverts: romantic music expressing overt emotions
Unstable introverts: mystical and impressionistic romantic works
Aggressive: heavy metal or hard rock
Japanese adolescents: classical or jazz
Critical age: mean 23.5 years old
Occupation
• Dressmakers: moderately slow
• Typist: fast tempo
– Socio-economic background
• Upper class women: classical
• Working class men: hillbilly (Indiana)
– Consistency in ranking of classical and popular music
– Enjoyment correlates to labeling (“romantic”, “Nazi”, none) or known
composer’s name
Uitdenbogerd, A., and R. Schyndel. 2002. A review of factors affecting music recommender
success. International Symposium on Music Information Retrieval. 204-8.
• Factors affecting music preference
–
–
–
–
–
–
–
–
–
–
Age
Origin
Occupation
Socio-economic background
Personality
Gender
Musical education
Familiarity with the music or style
Complexity of music
Lyrics
Uitdenbogerd, A., and R. Schyndel. 2002. A review of factors affecting music recommender
success. International Symposium on Music Information Retrieval. 204-8.
• Genres / styles
– AllMusicGuide.com: 531
– Amazon,com:
719
– MP3.com
430
• Moods
– 8 clusters with 67 moods (Hevner)
– 10 clusters with 52 moods (Farnsworth 1958)
– Features: tempo, tonality, distinctiveness of
rhythm, pitch height
Uitdenbogerd, A., and R. Schyndel. 2002. A review of factors affecting music recommender
success. International Symposium on Music Information Retrieval. 204-8.
Techniques for music recommenders
• Collaborative filtering
– Feedback from users: ratings, annotations, time spent
• Content-based filtering
–
–
–
–
–
Problem of extracting musical semantics from raw signal
Low-level features; notes, timbre, rhythm
High-level features: adjectives
Transcription, instrument identification, genre classifier
Similarity measure from user supplied example (Welsh et al.)
• 1248 features, 10-15 second samples, k-NN
Kim, J.-Y., and N. Belkin. 2002. Categories of music description and search terms
and phrases used by non-music experts. International Symposium on Music
Information Retrieval. 209-14.
• Information needs (music as information)
–
–
–
–
Information-seeking towards the satisfaction of user
Why does the user seek information?
What purpose does the user believe it will serve?
What use does it serve when found?
• Three basic “human needs”
– Physiological (food, water, shelter)
– Affective (emotional needs, e.g.: attainment, domination)
– Cognitive (need to plan, need to learn skills)
• Music IR has concentrated on cognitive needs
– Not enough user need studies
– Ignored affective needs
– Ignored musical information needs
Kim, J.-Y., and N. Belkin. 2002. Categories of music description and search terms
and phrases used by non-music experts. International Symposium on Music
Information Retrieval. 209-14.
• Purpose: To relate descriptions of affect to specific
musical works
– “means” for listeners to express their information “needs”
• Seven classical music: 22 subjects
– 11 s.: Words to describe the music
– 11 s.: Words used to search for the music
• Words used grouped into seven categories
– Mostly emotions and occasions or filmed events
• Subjects had no formal musical training
– Used non-formal music terms
– Terms not found in music query systems
Futrelle, J., and J. Stephen Downie. 2002. Interdisciplinary communities and
research issues in music information retrieval. International Symposium on Music
Information Retrieval. 215-21.
• Two main problems in MIR research
– No evaluation method
– Lack of user-need studies
• Overemphasis on research in QBH systems is
unsupportable given their doubtful usefulness
• Research into recommender systems common in
other domain is inexplicably rare
• Lack of user interface research
• Undue emphasis on Western music
Futrelle, J., and J. Stephen Downie. 2002. Interdisciplinary communities and
research issues in music information retrieval. International Symposium on Music
Information Retrieval. 215-21.
First Principles of MIR:
• MIR systems are developed to serve the needs of
particular user communities.
• MIR techniques are evaluated according to how well
they meet the needs of user communities.
• MIR techniques are evaluated according to agreedupon measures against agreed-upon collections of
data, so that meaningful comparisons can be made
between different research efforts.
Blandford, A., and H. Stelmaszewska. 2002. Usability of musical digital libraries:
A multimodal analysis. International Symposium on Music Information Retrieval.
231-7.
Evaluation of four web-accessible music libraries.
•
•
•
•
www.nzdl.org music
www.nzdl.org video
ABC Tunefinder
Folk Music Collection
•
Aimed at different user community (different levels of
technological and musical knowledge)
Too many file format choice for novices
•
Other usability studies
• Variations (Indiana Music Library)
• Design guidelines and user-centered digital libraries
(Theng et al.)