Markus Schedl and Tim Pohle

1
Exploring Music Artists via Descriptive
Terms and Multimedia Content
Markus Schedl and Tim Pohle Department of
Computational Perception Johannes Kepler
University Linz http//www.cp.jku.at
2
Topic of Talk
Suggest a new way to browse and access
multimedia content found on the web pagesof a
music artist
3
Overview

• Acquisition of Data offered for Browsing
• Suggested Approach The 3D Co-Occurrence
  Browser
• Evaluation A Building the Interface
• ? Different Term Weighting Functions
• Evaluation B The User Interface
• Integration of 3D COB in AGMIS
• Conclusions

4
Data Acquisition
musicreview
Alice Cooper BB King Beethoven Prince Met
allica
top-ranked URLs
Alice Cooper http//www.geocities.com/sfloman/alic
ecooperband.html http//music.yahoo.com/ar-307112-
reviews--Alice-Cooper http//music.yahoo.com/relea
se/165446 http//www.popmatters.com/music/reviews/
c/cooperalice-dirty.shtml http//www.popmatters.co
m/music/reviews/c/cooperalice-billion.shtml
BB King http//www.amazon.com/exec/obidos/tg/detai
l/-/B000AA4M9U?vglance http//www.amazon.com/exec
/obidos/tg/detail/-/B00004THAY?vglance http//www
.rollingstone.com/artists/4610/reviews http//www.
rollingstone.com/artists/4610/albums/album/7600591
http//www.popmatters.com/music/reviews/k/kingbb-
anthology.shtml
indexing
retrieve Web pages
store data
alternative banjo dirty rap gothic metal Joseph
Haydn
common file extensions for audio/image/video
(mp3, png, avi, )

• full inverted index
• XML structure

5
Browsing Artist-Related Web Pages Using
Sunbust-like Visualizations

• Data source full inverted file index
• Create co-occurrence tree and visualize it
• Algorithmic outline
• 1. Start at root node set of all Web pages
  retrieved for the artist
• 2. Select i most important terms ti (according
  to term weighting function)
• 3. Child nodes sets of Web pages containing
  terms t1, , ti
• 4. For each of these sets, goto 2. (until
  maximum depth reached)

6
All web pages for artist (e.g., for Iron Maiden)
Web pages for artist containing the term Metal
Web pages for artist containing the term Guitar

Web pages for artist containing the terms Metal
ANDGuitar


7
Sunburst / InterRing Visualization

• Circular, space-filling visualization technique
• Center represents root node
• Deeper elements in hierarchy are drawn further
  away from center
• Children are drawn within angular borders of
  their parent

8
Sunburst (2) How does it look like?
9
3D Co-Occurrence Browser

• Brings the Sunburst to 3D
• Additional data dimension can be encoded in
  height of each arc
• Stacking a number of such 3D-Sunbursts offers
  even more dimensions
• ? Amount of multimedia data found on the
  artist-related Web pages is visualized (three
  layers for audio, image, and video files)
• User interaction by rotating, zooming, changing
  the view angle, displaying Web pages and
  multimedia content

10
3D Co-Occurrence Browser (2)

• Constraints to limit the size of the Sunburst as
  before
• maximum number of terms with highest weights
• maximum depth of the tree
• minimum angular extent of each arc of the
  visualization

11
3D Co-Occurrence Browser (3)
12
Evaluation A How to Select Terms.

• Data source full inverted file index
• Evaluate different term weighting functions (TF,
  DF, TFxIDF) to rank terms from music dictionary
  occurring in corpus of artists Web pages.
• Question Which function selects the most
  meaningful terms to describe an artist?
• User study to assess descriptiveness of highest
  ranked terms, using the three different weighting
  functions
• 112 well known artists from 14 genres
• Web page indexing using dictionary of 1,506
  musically relevant terms
• 10 highest ranked terms of the 3 weighting
  functions merged
• ? 1 term set for each artist
• 5 participants, each told to rate terms for the
  artists they knew well(categorizing each term in
  three classes , -, )

13
Evaluating Different Term Weighting Functions (2)

• 172 individual artist ratings returned
• 92 of 112 artists covered
• Overall excess of good terms () over bad terms
  (-)
• TF 2.22
• DF 2.42
• TFxIDF 1.53
• TF and DF performed significantly better than
  TFxIDF, no significant difference between TF and
  DF

14
Evaluation B The User Interface

• Explaining the functionality of the 3D COB to the
  participants
• Warm-up phase (participants were invited to play
  around with the 3D COB of Britney Spears)
• Evaluation carried out on 3D COB of Iron Maiden
• Different tasks to be solved by the participants
• Tasks addressing the co-occurring terms
• 1) Which are the 5 top-ranked terms that occur
  on the web pages mentioning Iron Maiden?
• 2) Indicate the number of web pages containing
  all of the terms Iron Maiden, metal, and
  guitar.
• 3) Show a list of web pages that contain the
  terms Iron Maiden and british.
• 4) Considering the complete set of web pages,
  which are the 3 terms that co-occur on the
  highest number of web pages?
• 5) How many web pages contain the terms Iron
  Maiden and metal, but not the term guitar?

15
Evaluating the User Interface (2)

• Different tasks to be solved by the participants
• Tasks addressing the dimension of multimedia
  content
• 6) Display a list of audio files available at
  web pages containing the term Iron Maiden.
• 7) Which terms co-occur on the set of web pages
  that contains the highest number of image files
  in hierarchy level 3?
• 8) Indicate the URL of one web page that
  contains image files but no video files.
• Tasks related to the structure of the Sunburst
  tree
• 9) How many web pages does the complete
  collection contain?
• 10) Find one of the deepest elements in the
  hierarchy and select it.
• 11) Generate a new visualization using only the
  web pages on which the terms bass and heavy
  metal co-occur.
• Time to complete each task measured
• 6 participants (computer science and business
  students, 5 male)

16
Evaluating the User Interface (3)

• Tasks related to structural questions were
  answered in a shorter time than those requiring
  browsing the collection.
• Intensive rotation requirements hindered fast
  completion of tasks.
• ? automated rotation to selected arc !
• High average time to perform task 1 probably due
  to unfamiliar new visualization of Iron Maiden.
• Average time to complete a task 45 seconds
• Promising results in comparison to other tree
  visualizations systems, but user interaction
  functionalities require improvements.

17
Building an Automatically Generated MIS
Similar Artists
Prototypical Artist Detection
Band Member and Instrumentation Detection
UI to Browse Artist-Related Web Pages
Automatic Attribution / Tagging
Album Cover Retrieval
18
Use of the 3D-COB in AGMIS
19
Conclusions

• 3D Co-Occurrence Browser as extension of the
  Sunburst visualization technique
• Data acquired from the Web, indexing via
  musically relevant terms and MM file extensions
• Term weighting DF and TF outperformed TFxIDF in
  quality of descriptive terms
• Assessment of UI showed that 3D COB offers
  additional information that cannot be discovered
  by the standard list-based search results offered
  by common Web search engines
• 3D COB successfully integrated into an
  automatically generated music information system
  (AGMIS) more details in
• M. Schedl, PhD thesis Automatically
  Extracting, Analyzing, and Visualizing
  Information on Music Artists from the World Wide
  Web, 2008. http//www.cp.jku.at/people/schedl/

20
Thank you.