Active Mobile Music Discovery
with Social Tags
Torsten Möller
IUI 2016
March 2016, Sonoma, CA
Mohsen
Kamalzadeh
Christoph ...
motivation design evaluation conclusion
• online streaming services
• over 30 millions songs
• need 200 years to listen to...
motivation design evaluation conclusion
play something nice for
a while
exploration and
discovery
3
how to narrow down?
motivation design evaluation conclusion
play something nice for
a while
exploration and
discovery
4
• radio stations
• cur...
motivation design evaluation conclusion
play something nice for
a while
exploration and
discovery
5
• radio stations
• cur...
motivation design evaluation conclusion
?6
play something nice for
a while
• radio stations
• curated playlists
• recommen...
motivation design evaluation conclusion
?7
play something nice for
a while
• radio stations
• curated playlists
• recommen...
motivation design evaluation conclusion
?8
play something nice for
a while
• radio stations
• curated playlists
• recommen...
motivation design evaluation conclusion
• elaborate
interfaces
• large screens
• for long sessions
?9
play something nice ...
motivation design evaluation conclusion
• elaborate
interfaces
• large screens
• for long sessions
?10
play something nice...
motivation design evaluation conclusion
• elaborate
interfaces
• large screens
• for long sessions
?11
play something nice...
motivation design evaluation conclusion
• elaborate
interfaces
• large screens
• for long sessions
?12
play something nice...
motivation design evaluation conclusion
?13
play something nice for
a while
• radio stations
• curated playlists
• recomme...
motivation design evaluation conclusion
14
play something nice for
a while
• radio stations
• curated playlists
• recommen...
motivation design evaluation conclusion
15
TagFlip
high user control
+
low interaction effort
+
small screen
motivation design evaluation conclusion
TagFlip
Contributions:
• Design and evaluation of a tag-based
interactive recommen...
motivation design evaluation conclusion
17
Design
motivation design evaluation conclusion
• High level abstraction over music library
• Understandable to ordinary users
• N...
motivation design evaluation conclusion
19
motivation design evaluation conclusion
20
• High impact interaction
• Fine tuning/precise control
• Low interaction effor...
motivation design evaluation conclusion
21
• left to right flow
design choices
Iterative user-centred design
motivation design evaluation conclusion
22
• left to right flow
design choices
Iterative user-centred design
motivation design evaluation conclusion
23
• left to right flow
• tag strength
design choices
Iterative user-centred design
motivation design evaluation conclusion
24
• left to right flow
• tag strength
design choices
Iterative user-centred design
motivation design evaluation conclusion
25
• left to right flow
• tag strength
• communicating target set sizes
design cho...
motivation design evaluation conclusion
26
• left to right flow
• tag strength
• communicating target set sizes
• “excludi...
motivation design evaluation conclusion
27
Evaluation
motivation design evaluation conclusion
• Is there room for this type of interface?
• compared with Spotify’s mobile app
•...
motivation design evaluation conclusion
29
• with each app:
- alternating order of apps
- 5 minutes familiarization
- 10 m...
motivation design evaluation conclusion
30
questionnaires video captureinterview
evaluation data
motivation design evaluation conclusion
31
questionnaires video captureinterview
evaluation data
• Usability
- SUS questio...
motivation design evaluation conclusion
32
video captureinterview
evaluation data
• Usability
- SUS questionnaire
• Recomm...
motivation design evaluation conclusion
33
video capture
evaluation data
• Objective
performance
interview
• Pros and cons...
motivation design evaluation conclusion
34
Hypotheses
• H1: Higher overall score for TagFlip in rec. questionnaire
• H2: S...
motivation design evaluation conclusion
1.Quality of
recommendations
35
questionnaires
• Usability
• Recommendation
video ...
motivation design evaluation conclusion
0 1 2 3 4
2.67
2.26
2.33
2.29
2.51
3.03
2.83
2.72
2.48
3.03
TagFlip Spotify
36
que...
motivation design evaluation conclusion
0 1 2 3 4
2.67
2.26
2.33
2.29
2.51
3.03
2.83
2.72
2.48
3.03
TagFlip Spotify
37
que...
motivation design evaluation conclusion
38
video capture
interview
• Objective
performance
10 20 30 40
34.58
25.61interact...
motivation design evaluation conclusion
39
video capture
interview
• Objective
performance
10 20 30 40
34.58
25.61interact...
motivation design evaluation conclusion
video capture
interview
40
• pros:
• precise control (all 16
participants)
• fine-...
motivation design evaluation conclusion
video capturequestionnaires interview
• Higher overall satisfaction with TagFlip
•...
motivation design evaluation conclusion
42
Conclusion
motivation design evaluation conclusion
43
• stagnant mobile recommendation scene
• encouraging results with TagFlip
• gre...
motivation design evaluation conclusion
44
• design solutions to remaining problems
• excluding tags, tag strength, etc.
•...
motivation design evaluation conclusion
45
Thank you!
Upcoming SlideShare
Loading in …5
×

TagFlip: Active Mobile Music Discovery with Social Tags

694 views

Published on

The presentation for the TagFlip paper at ACM IUI 2016 conference.
TagFlip was built for phone-sized screens, and couples high user control on the recommended music with minimal interaction effort. Contrary to conventional recommenders, which only allow the specification of seed attributes and the subsequent like/dislike of songs, we put the users in the centre of the recommendation process. With a library of 100,000 songs, TagFlip describes each played song to the user through its most popular tags on Last.fm and allows the user to easily specify which of the tags should be considered for the next song, or the next stream of songs. In a lab user study where we compared it to Spotify’s mobile application, TagFlip came out on top in both subjective user experience (control, transparency, and trust) and our objective measure of number of interactions per liked song. Our users found TagFlip to be an important complementary experience to that of Spotify, enabling more active and directed discovery sessions as opposed to the mostly passive experience that traditional recommenders offer.

Paper: http://dl.acm.org/citation.cfm?doid=2463656.2856780&conferenceID=139

Published in: Science
0 Comments
1 Like
Statistics
Notes
  • Be the first to comment

No Downloads
Views
Total views
694
On SlideShare
0
From Embeds
0
Number of Embeds
7
Actions
Shares
0
Downloads
2
Comments
0
Likes
1
Embeds 0
No embeds

No notes for slide

TagFlip: Active Mobile Music Discovery with Social Tags

  1. 1. Active Mobile Music Discovery with Social Tags Torsten Möller IUI 2016 March 2016, Sonoma, CA Mohsen Kamalzadeh Christoph Kralj Michael Sedlmair TagFlip
  2. 2. motivation design evaluation conclusion • online streaming services • over 30 millions songs • need 200 years to listen to everything! • how to narrow down? 2
  3. 3. motivation design evaluation conclusion play something nice for a while exploration and discovery 3 how to narrow down?
  4. 4. motivation design evaluation conclusion play something nice for a while exploration and discovery 4 • radio stations • curated playlists • recommendation how to narrow down?
  5. 5. motivation design evaluation conclusion play something nice for a while exploration and discovery 5 • radio stations • curated playlists • recommendation • library visualization • critiquing, weighting, etc. how to narrow down?
  6. 6. motivation design evaluation conclusion ?6 play something nice for a while • radio stations • curated playlists • recommendation exploration and discovery • library visualization • critiquing, weighting, etc.
  7. 7. motivation design evaluation conclusion ?7 play something nice for a while • radio stations • curated playlists • recommendation exploration and discovery • library visualization • critiquing, weighting, etc. • not much control • black box problem • pigeon-holing
  8. 8. motivation design evaluation conclusion ?8 play something nice for a while • radio stations • curated playlists • recommendation exploration and discovery • library visualization • critiquing, weighting, etc. • not much control • black box problem • pigeon-holing • elaborate interfaces • large screens • for long sessions
  9. 9. motivation design evaluation conclusion • elaborate interfaces • large screens • for long sessions ?9 play something nice for a while • radio stations • curated playlists • recommendation exploration and discovery • library visualization • critiquing, weighting, etc. • not much control • black box problem • pigeon-holing Pu and Chen, 2006
  10. 10. motivation design evaluation conclusion • elaborate interfaces • large screens • for long sessions ?10 play something nice for a while • radio stations • curated playlists • recommendation exploration and discovery • library visualization • critiquing, weighting, etc. • not much control • black box problem • pigeon-holing Pampalk and Goto, 2007
  11. 11. motivation design evaluation conclusion • elaborate interfaces • large screens • for long sessions ?11 play something nice for a while • radio stations • curated playlists • recommendation exploration and discovery • library visualization • critiquing, weighting, etc. • not much control • black box problem • pigeon-holing O’Donovan et al., 2008
  12. 12. motivation design evaluation conclusion • elaborate interfaces • large screens • for long sessions ?12 play something nice for a while • radio stations • curated playlists • recommendation exploration and discovery • library visualization • critiquing, weighting, etc. • not much control • black box problem • pigeon-holing Bostandjiev et al., 2012
  13. 13. motivation design evaluation conclusion ?13 play something nice for a while • radio stations • curated playlists • recommendation exploration and discovery • library visualization • critiquing, weighting, etc. • not much control • black box problem • pigeon-holing • elaborate interfaces • large screens • for long sessions
  14. 14. motivation design evaluation conclusion 14 play something nice for a while • radio stations • curated playlists • recommendation exploration and discovery • library visualization • critiquing, weighting, etc. • not much control • black box problem • pigeon-holing TagFlip • elaborate interfaces • large screens • for long sessions
  15. 15. motivation design evaluation conclusion 15 TagFlip high user control + low interaction effort + small screen
  16. 16. motivation design evaluation conclusion TagFlip Contributions: • Design and evaluation of a tag-based interactive recommendation interface • Design choices, lessons, and future directions • Strengths of tag-based discovery 16
  17. 17. motivation design evaluation conclusion 17 Design
  18. 18. motivation design evaluation conclusion • High level abstraction over music library • Understandable to ordinary users • Navigate based on social tags • genre • mood • location • year • instruments • etc. • 100,000 songs • 350 tags (processed from 500,000) • Android 18 TagFlip
  19. 19. motivation design evaluation conclusion 19
  20. 20. motivation design evaluation conclusion 20 • High impact interaction • Fine tuning/precise control • Low interaction effort and mental load • Transparency • Small screen key design factors
  21. 21. motivation design evaluation conclusion 21 • left to right flow design choices Iterative user-centred design
  22. 22. motivation design evaluation conclusion 22 • left to right flow design choices Iterative user-centred design
  23. 23. motivation design evaluation conclusion 23 • left to right flow • tag strength design choices Iterative user-centred design
  24. 24. motivation design evaluation conclusion 24 • left to right flow • tag strength design choices Iterative user-centred design
  25. 25. motivation design evaluation conclusion 25 • left to right flow • tag strength • communicating target set sizes design choices Iterative user-centred design
  26. 26. motivation design evaluation conclusion 26 • left to right flow • tag strength • communicating target set sizes • “excluding” tags design choices Iterative user-centred design
  27. 27. motivation design evaluation conclusion 27 Evaluation
  28. 28. motivation design evaluation conclusion • Is there room for this type of interface? • compared with Spotify’s mobile app • within subjects • 16 users (8 female), median age = 26 28 Evaluation
  29. 29. motivation design evaluation conclusion 29 • with each app: - alternating order of apps - 5 minutes familiarization - 10 minutes test - find new songs that you like - use any feature of apps - usability and recommendation questionnaires • post-hoc interview Protocol
  30. 30. motivation design evaluation conclusion 30 questionnaires video captureinterview evaluation data
  31. 31. motivation design evaluation conclusion 31 questionnaires video captureinterview evaluation data • Usability - SUS questionnaire • Recommendation • Based on ResQue framework (Pu and Chen, RecSys 2010)
  32. 32. motivation design evaluation conclusion 32 video captureinterview evaluation data • Usability - SUS questionnaire • Recommendation • Based on ResQue framework (Pu and Chen, RecSys 2010) questionnaires • Pros and cons of each app
  33. 33. motivation design evaluation conclusion 33 video capture evaluation data • Objective performance interview • Pros and cons of each app • Usability - SUS questionnaire • Recommendation • Based on ResQue framework (Pu and Chen, RecSys 2010) questionnaires
  34. 34. motivation design evaluation conclusion 34 Hypotheses • H1: Higher overall score for TagFlip in rec. questionnaire • H2: Smaller number of interactions per liked song in TagFlip • H3: More liked songs with TagFlip questionnaires video captureinterview evaluation data
  35. 35. motivation design evaluation conclusion 1.Quality of recommendations 35 questionnaires • Usability • Recommendation video captureinterview 2.Interface and interaction adequacy 3.Control and transparency 4.Attitudes and behavioural intentions (trust)
  36. 36. motivation design evaluation conclusion 0 1 2 3 4 2.67 2.26 2.33 2.29 2.51 3.03 2.83 2.72 2.48 3.03 TagFlip Spotify 36 questionnaires 2.Interface and interaction adequacy 1.Quality of recommendations 3.Control and transparency 4.Attitudes and behavioural intentions (trust) scores from 0 to 4 video captureinterview p values form paired sample t-test questionnaires • Usability • Recommendation p<0.05 All questions p<0.05 p<0.05 p<0.05
  37. 37. motivation design evaluation conclusion 0 1 2 3 4 2.67 2.26 2.33 2.29 2.51 3.03 2.83 2.72 2.48 3.03 TagFlip Spotify 37 questionnaires 2.Interface and interaction adequacy 1.Quality of recommendations 3.Control and transparency 4.Attitudes and behavioural intentions (trust) scores from 0 to 4 video captureinterview • TagFlip sig. higher in • overall subjective recommendation capability (Hypothesis 1 confirmed) p values form paired sample t-test questionnaires • Usability • Recommendation p<0.05 All questions p<0.05 p<0.05 p<0.05
  38. 38. motivation design evaluation conclusion 38 video capture interview • Objective performance 10 20 30 40 34.58 25.61interactions per liked song 2 4 6 8 10 7.06 7 number of liked songs questionnaires p<0.05
  39. 39. motivation design evaluation conclusion 39 video capture interview • Objective performance 10 20 30 40 34.58 25.61interactions per liked song 2 4 6 8 10 7.06 7 number of liked songs questionnaires p<0.05 • TagFlip needed sig. less interaction (H2 confirmed) • Equal number of liked songs (H3 not confirmed
  40. 40. motivation design evaluation conclusion video capture interview 40 • pros: • precise control (all 16 participants) • fine-tuning (11) • cons: • Spotify better for passive listening (8) • 7 participants very enthusiastic • 7 other participants open to using questionnaires
  41. 41. motivation design evaluation conclusion video capturequestionnaires interview • Higher overall satisfaction with TagFlip • High user control with TagFlip • Lower interaction effort with TagFlip • Fine-tuning ability is key in user enthusiasm 41
  42. 42. motivation design evaluation conclusion 42 Conclusion
  43. 43. motivation design evaluation conclusion 43 • stagnant mobile recommendation scene • encouraging results with TagFlip • great potential for tag-based interaction Conclusion
  44. 44. motivation design evaluation conclusion 44 • design solutions to remaining problems • excluding tags, tag strength, etc. • longitudinal studies • more advanced processing • excluding music types? • topic modelling Future work
  45. 45. motivation design evaluation conclusion 45 Thank you!

×