Interactive Exploration of
Music Listening Histories
Ricardo Dias, Manuel J. Fonseca,
Daniel Gonçalves
Context & Motivation
Proliferation of lifelogging services
Tracking Listening Habits
Music Listening Histories
User 1
  Artist 1, Song 1, Timestamp
  Artist 2, Song 3, Timestamp
  Artist 1, Song 2, Timestamp   Time
  Artist 3, Song 5, Timestamp
  Artist 4, Song 4, Timestamp
  …
User 2
  …
Profile Generation
Recommendation
Pattern detection in user habits




                             zachstern@flickr
Good at recording data…
… and allow direct access to it!
But Visualization is also important!




                              ottonassar@flickr
Related work
StreamGraph [Byron and Wattenber 2008]
Tangle, Strings and Knots [Baur2009]
LastHistory [Baur2011]
Fans effort to visualize data
Scrobbling Timeline                    Arc diagrams




    LastGraph
Problems
Main issues
o More concerned about design and aesthetics
o Static visualizations and only overviews of
  listening histories
o Lack of interactive browsing and filtering
o Scalability issues, regarding the number of
  songs to represent
Main issues
o More concerned about design and aesthetics
o Static visualizations and only overviews of
  listening histories
o Lack of interactive browsing and filtering
o Scalability issues, regarding the number of
  songs to represent
Main issues
o More concerned about design and aesthetics
o Static visualizations and only overviews of
  listening histories
o Lack of interactive browsing and filtering
o Scalability issues, regarding the number of
  songs to represent
Main issues
o More concerned about design and aesthetics
o Static visualizations and only overviews of
  listening histories
o Lack of interactive browsing and filtering
o Scalability issues, regarding the number of
  songs to represent
Our solution
Design principles

Rationale
Exploration




              thegiantvermin@flickr
Overview




           [Dias and Fonseca2010]
Interactivity
                Ultrastart
Details on demand




                    m4tik@flickr
Inference




            mterraza@sxc
Analysis
Music Listening History Explorer

MULHER
MULHER
Main Visualization: Stacked dot
Main Visualization: Timeline
Main Visualization: Background
Filtering
o Genre
o Artist name    Single selection

o Free Text
                                    Multiple Selection
o Time


                 Combining Filters
Statistics
General Statistics     Daily and Hourly Visualizations
                       With Context   Without Context
Brushing & Highlighting
Implementation
Regular Web Application   Backend Server
 HTML, CSS, Javascript
      Protovis             Java backend
 JSON to represent data
Demo
Evaluation
Two Experiments




                  swamibu@flickr
First experiment

Common listening history
Evaluate:
  – UI usability
  – Users satisfaction
    and experience
Second Experiment: Case Studies

o Users exploring their own listening histories
o Listening Patterns Detection
Experiments Details


Test Procedure         Users
     Tasks        Listening Logs


     =                 ≠
Setup Procedure
o 45 min for each test
o User’s personal computers




                              mytudut@flickr
Test Session Procedure
1.   Quick introduction
2.   Application description
3.   Practice Time
4.   Tasks execution
5.   Satisfaction survey & Informal Interview
Tasks
9 tasks
2 Categories
  1. Explore and browse the listening history
  2. Pattern detection and trends on music listening
     habits
Exploring Tasks: examples
Indicate the most played song of the artist X
over the last three months

Describe the trend on the previous identified
song
Exploring Tasks: examples
Indicate the most played song of the artist X
over the last three months

Describe the trend on the previous identified
song
Pattern Detection Tasks: examples
Describe and try to justify the listening changes
that occurred over the last three months

Describe the listening habits on the selected
period
Pattern Detection Tasks: examples
Describe and try to justify the listening changes
that occurred over the last three months

Describe the listening habits on the selected
period
First Experiment
Common Log
o Listening Histories from one of the authors
o 5.000 records (11/2009 to 07/2011)




                                           alaasafei@sxc
Participants

                          8        2

N = 10
Ages between 20-50 years
Listen to music almost every day
Occupations

4 CS students
2undergraduate CS students
  2 SW engineers
         1 journalist
1 HS student
Statistical Evaluation
o Effectiveness
o Error rate
o Satisfaction




                                   iamwahid@sxc
First Experience

Results
Effectiveness




100%
Overall Task Success Rate




  97%               3%
Successfull      Unsuccessfull
Problem of Task 3
# Users
   10


                    6



   0
          T1   T2   T3   T4   T5   T6   T7   T8   T9
Overall Tasks Easiness




                           1      8         1
  0             0
 Very        Difficult   Normal   Easy   Very easy
difficult
User Experience Satisfaction
    Fantastic   Rewarding     Stimulant   Easy   Flexible
9
8
7
6
5
4
3
2
1
    Horrible    Frustrating    Boring     Hard   Rigid
Second Experiment
Users’ logs
o Personal listening histories
o From 3.000 to 30.000 records




                                 alaasafei@sxc
Participants: Second Experience



                      #N = 4
Ages between 20-40 years
All listen to music every day
Active Last.fm account
Occupations


2 CS Assistant Teachers
  1 Senior Software Engineer

     1 CS PhD Student
Results
Identical Statistical Results




                          lumaxart@flickr
Pattern Inference – Users interviews




                                whitebeard@sxc
Life Past Events
User 1
“Here I was working on a scientific paper,
because I was listening only classical music, and
I like to hear that kind of music when I'm
writing, But then I skipped listening to music,
because I had some project discussions, and no
time to listen to music”
Profile Inference
User 2
"Well, do you know why this part of the
visualization contains mostly recent music, even
though I just prefer to listen old music?“
Profile Inference
User 2
"Well, do you know why this part of the
visualization contains mostly recent music, even
though I just prefer to listen old music?“

 one of his favorite artists just released a
new album after years of absence
Hidden Time Habits
User 4
"I did not realized that I was listening too much
music in late night, but now that I think of this, I
usually listen to more rhythmic music at that
time to stay awake a little longer, mostly when
I'm working"
Listening Trends
User 3
"Looks like that through a regular day I keep
changing the genre of music I listen to. I start
with something stronger in the morning and
then end the day with more relaxing songs!"
Discussion
Timeline
o Timeline-based mechanism proved to be a
  major asset:
  – Main browsing and filtering technique
  – Effectiveness and flexibility validated by
    experimental results
Context Information
o Considered to be an important aspect of the
  solution
  – Info about most played elements
  – Acts as a visual clue to start the exploration
“Age of Songs”
o Can effectively convey information about the
  listening habits
  – Different profiles by direct color inspection
Knowledge analysis and inference
o Performed by combining insights from the
  different techniques

o Possible main based on time, absence of
  music listening and context information
Conclusions
Conclusions
Novel solution for exploring and filtering
listening histories
  1. Combines a timeline-based visualization with a
     set of synchronized-views to perform direct
     exploration
  2. Introduces a new feature: the “age” of songs
  3. Allows listening pattern detection, only based on
     time and some textual metadata
Future Work
o Data mining on listening histories data
  – Discover new hidden listening patterns
  – New pattern examples:
     • Users that always seek the tops
     • Others that enjoy mostly female voices
     • Listen to classic music in the morning but rhythmic
       music in the afternoon
o Map the new patterns to this and other
  visualizations
Thank you
Questions?



ricardo.dias@ist.utl.pt
web.ist.utl.pt/~ricardo.dias
MULHER@AVI2012

MULHER@AVI2012