1. Survey Of Music Information Needs, Uses, and Seeking Behaviors Jin Ha Lee J. Stephen Downie Graduate School of Library and Information Science University of Illinois at Urbana-Champaign

2. HUMIRS Project Human Use of Music Information Retrieval Systems Sub-project of the International Music Information Retrieval Systems Evaluation Laboratory (IMIRSEL) To provide real-world grounding for the evaluation tests To help us form the standardized query documents

3. Characteristics of the subjects 437 responses from the UIUC campus population Top-ranked genres: Rock, Pop, Classical and Alternative Avid listeners (73.1%), Musically passionate (36.6%) 63.6% can read sheet music “OK” to “very well” 64% expressed their singing ability is average or above 74.5% can play a musical instrument

4. Findings 1. Descriptive metadata and extra-musical information have important commercial and experience enrichment aspects for users. 2. Users seek music as an auditory experience. 3. Users seek music information to assist in building collections of music. 4. Users seek music information for verifying or identifying works, artists, lyrics, etc. 6. Users prefer online resources for extra-musical information. 7. Users have definite preferences regarding where they physically seek music information.

5. Findings 5. Users value online music reviews, ratings, recommendations, and suggestions. 8. Personal familiarity with search helpers is a key determinant for music information seekers. 9. Music information-seeking should be seen as a socially instigated act. 10. Music information seekers employ public knowledge and/or opinions for searches.

6. Public information-seeking People are generating or using collective knowledge in their music information-seeking More flexible and less directed process of exploration Social network has a major impact Positive opinions towards recommendations from other people, reviews, ratings, etc. (i.e., extra-musical information) Amazon.com

7. Supportive data Persons consulted for search Friends and Family Members (84%) (29.9% - more than once a week) Places visited for search Friend’s or Acquaintance’s Place (76%) (18.2% - more than once a week) Sources that triggered search Friend’s or Acquaintance’s Place (87.1%)

8. Supportive data Search or browse by Recommendations from other people (62.1%)58.9%) Likely to seek Reviews and rating by others (49.1%)(49.7%)

9. Need for context metadata Positive opinions toward extra-musical data Relations: Genre, Similar music, Similar artist Associated use: searches triggered by Radio show, TV show, movie or animation, advertisement or commercial Allmusic.com “I learn something new every time I go there.”

10. Supportive data Sources that triggered search Radio show (81.4%) (82.5%) TV show, Movie, Animation (80.8%) (65%) Advertisement/Commercial (63.8%) (53.1%) Search or browse by Similar artists (59.4%) (52.5%) Similar music (54.5%) (47.5%) Associated use (41.7%) (31.8%)

11. Types of metadata Content Metadata Musical metadata: data derived directly from the music itself (e.g., melody, tempo, etc.) Bibliographic metadata: traditionally-used metadata that describes the item (e.g., title, author, etc.)

12. Types of metadata Context Metadata Relational metadata: data about the item’s relationships (artificially created or socially constructed) with other music related items (e.g., genre; indications of similarity, etc.) Associative metadata: data indicating associated use in other works, media or events (e.g., sampling; use in TV show, movies or commercials; use at special events, etc.)

13. Example ContentMusicalSample track BibliographicTitle: Thank you Singer: Dido ContextRelationalGenre: Rock Styles: Adult/Alternative pop AssociativeUse: Theme of the TV show “Roswell” Related work: Sampled in “Stan” by Eminem

14. How to collect context metadata? Hard to generate automatically Cannot be generated solely from an individual item or at the point of the item’s production/creation One possible way is to include music community members or subject enthusiasts Research in progress

15. Future research Further statistical analysis of the survey data Correlation between multiple variables Web music queries analysis Google answers, www.musicsearch.comwww.musicsearch.com Markup framework for music queries

16. Questions? Comments? Special Thanks to Andrew W. Mellon Foundation National Science Foundation