Smart Party: Song Selection

1
Smart Party Song Selection
Kevin F. Eustice, Jason Stoops, Peter Reiher,
Venkatraman Ramakrishna, Nam Nguyen
UCLA
2
Voting Algorithm Overview (Eustice)?

• Each device nominates a song to be played next
• A ballot containing metadata of all these songs
  is sent out to all devices
• From a pool of points, devices then assign points
  to each song in ballot proportional to personal
  preference value associated with song
• Points are tallied by coordinator and the bottom
  half of candidates are dropped
• Process repeats until one song wins!

3
Point Assignment

• Matches attempted by devices for point assignment
  when filling out ballot
• Filename match --gt points based on song
  preference
• Artist, title match --gt points based on song
  preference
• Artist match --gt based on average preference for
  artist
• Genre match --gt based on average preference for
  that genre
• No match --gt no points allocated to this
  candidate
• Averages for known genres and artists are
  computed at application startup

4
Ballot Compilation

• Logical ballot compilation ensures correct point
  distribution during vote
• Songs nominated by two different devices --gt one
  ballot entry
• Songs with same metadata but different filename
  --gt one ballot entry
• Songs with empty metadata --gt separate ballot
  entries.
• Devices can look at history of songs played in
  room and songs heard by device, and nominate
  unheard songs. (LASR)?

5
Robustness

• Vote coordination is done by application separate
  from Smart Party Location Application, or SPLA.
• SPLA has had reliable Round Robin song selection
  algorithm built in since beginning (LASR).
• If something goes wrong in voting, SPLA can
  detect this condition and fails over to Round
  Robin selection.
• Smart Party User Device Application, or SPUDA, is
  selection algorithm agnostic.
• simply responds to various requests

6
Accomplishments for quarter

• Implement voting algorithm
• Learn basics of Panoply framework
• Build up side-by-side implementation of voting
• Initially entirely separate from Smart Party
  application
• Integrate voting into SPUDA
• Vote Leader is still separate for future work in
  Smart Party
• Vote coordination may eventually be handled by
  SPUDAs
• Vote leader may be chosen dynamically in future

7
Accomplishments for quarter, cont.

• Matching for metadata and tracking average
  preference value for known artists, genres
• Fail over to Round Robin selection if vote
  coordinator dies for some reason
• Generally complete implementation of individual
  voting song selection for a multi-room Smart
  Party.

8
An interesting test

• Tested voting algorithm with 6 participants
• Varied media libraries and preferences
• Four people liked various songs by Artist A
• Not all people had all songs by artist, some had
  exclusive sets of songs.
• Two other people had a CD by another Artist B
• Overall, most music fell under Genre A.
• Result
• Songs by Artist A in Genre A played first, then
  other songs in Genre A.

9
Satisfaction Of Proposed Goals

• Satisfied Checkpoint 1 implement voting
  algorithm in smart party
• Algorithm is in use in 3 room demo in LASR lab
• Laptops can connect, and party will select songs
  from users in each room.
• However, checkpoint 2 has changed significantly...

10
Next quarter... Checkpoint 2

• Formulate user mobility model
• How does music affect people moving between rooms
  in the party?
• How does this affect fairness and user
  satisfaction of song selection algorithms?
• Investigate media-centric rooms vs. user-centric
  rooms
• Look at ALL music in party vs. music in room
• Create rooms based on preference groups
• Rock room, Jazz room, Beatles room, etc.
• Devices may suggest a room to user based on prefs

11
Acknowledgements

• Venkatraman Ramakrishna, UCLA LASR, for helping
  and guiding me during this project.
• Kevin F. Eustice, Nam Nguyen, and Peter Reiher,
  UCLA LASR