View on GitHub

MusEmoji

A Music Suggestion System Based Upon Emoji Selection

Download this project as a .zip file Download this project as a tar.gz file

Paper:

Download PDF

Our Project

There exist many ways to classify associated music—from quantitative audio characteristics to qualitative genre classifications. Our project sought to incorporate Emojis, a new component of modern electronic communication, into these systems of classification and association.
Please download our Paper to learn more about project!

Data Collection Results

  1. "Since I Met You Baby" by Kenny Neal
  2. "Just a Little Bit of You" by Michael Jackson
  3. "If I Ain't Got You" by Alicia Keys
  4. "Forever" by Chris Brown
  5. "She Will Be Loved" by Maroon 5
  6. "Let's Make Love (Album)" by Faith Hill
  7. "Shape Of My Heart" by Backstreet Boys
  8. "Fall In Love (Rub A Dub Mix)" by John Legend
  9. "My Baby" by Janet Jackson
  10. "Piece of My Heart" by Janis Joplin
  11. "Raining On Our Love" by Shania Twain
  12. "Will You Love Me Tomorrow" by The Shirelles
  13. "Permanent Lonely" by Billie Jo Spears
  14. "Don't Worry Baby" by Bryan Ferry
  15. "Be With Me" by Mindy McCready
  16. "I Only Have Eyes For You" by John Stevens
  17. "Don't Go Away" by Oasis

  1. "Fake Plastic Trees (Acoustic)" by Radiohead
  2. "Karma" by Alicia Keys
  3. "Paralyzed" by The Cardigans
  4. "Soot and Stars" by The Smashing Pumpkins
  5. "Losin' End (LP Version)" by Michael McDonald
  6. "Beautiful" by Eminem

  1. "Heart-Shaped Box" by Nirvana
  2. "Spin (Album Version)" by Taking Back Sunday
  3. "Metal" by Nine Inch Nails
  4. "Slaughter The Gods" by The Accursed
  5. "The Heretic Anthem (Live)" by Slipknot
  6. "Bite Back" by The All-American Rejects
  7. "4 Words (To Choke Upon)" by Bullet For My Valentine
  8. "Lesson Learned" by Alice In Chains

  1. "Let's Get It Started" by Black Eyed Peas
  2. "Touch The Sky" by Kanye West
  3. "Don't Stop The Music" by Rihanna
  4. "Into The Groove" by Madonna
  5. "Change Clothes" by Jay-Z
  6. "Hips Don't Lie" by Shakira
  7. "Poker Face" by Lady Gaga
  8. "Lively Up Yourself" by Bob Marley
  9. "Heaven's In New York" by Wyclef Jean

Classification Examples

A Clear Classification

"Let's Make Love" by Faith Hill

Classification: 19 - 1 - 0 - 0

This results of this classification were expected and displayed a clear association with listeners and the above Emoji.

An Unclear Classification

"Raining On Our Love" by Shania Twain

Classification: 10 - 9 - 1 - 0
OR
This classification was unclear as it was nearly evenly split between the two Emojis above--ten out of twenty manual classifications pointed to the former and nine out of the twenty pointed to the latter. We did anticipate this divide as the themes of the song are both love and sadness. However, this provides a clear example of two things: First, the music is subjective, as some can listen to a song and associate with one thing while another person can associate said song with something entirely different. Second, music is multidimensional and can represent multiple things at one time.

An Unexpected Classification

"Piece Of My Heart" by Janis Joplin

Classification: 11 - 3 - 5 - 1
Expected:
OR
Data Results:

We were surprised by this classification. While we did expect that this song could have different Emoji associations for different people, we anticipated that there would be a strong association by participants with the two expected Emoji. However, most people actually associated this song with the heart-eyes Emoji.
While we anticipated the upset tone of the lyrics and vocals to dominate for listeners and lead them to choose our expected Emojis, this was not the case. While eight total listeners did behave as anticipated and choose one of the two Emojis we expected, the majority (11) chose the heart-eyes Emoji. One possible explanation for this is that while the song does focus on a theme of being upset, it is about a former lover and, thus, may have swayed listeners to classify it with the heart-eyes Emoji.
Similar to the "Unclear Classification" example above, this data exemplifies the subjectivity of music in that different people perceive different things when listening to the same music. Furthermore, its wide distribution of "votes" across the four Emojis also point to the multidimensionality of music in that music can represent multiple themes concurrently.

Building a Nearest Neighbor Classifier

After organizing the data, the next step was to create a system similar to a nearest neighbor classifier. Using C++, we created a program that will take in the Mechanical Turk Results from a text file, then take in "test inputs" from a text file. It then compares each test song with each song from the Mechanical Turk Data in order to calculate which song is the closest. In the end, when a new song needed to be classified, this system would have told us which song, that was already classified, it was closest to.

Suggested Songs


"One Fine Day" by The Chiffons


"Angeles" by Elliott Smith


"The Plot To Bomb The Panhandle" by A Day To Remember


"London Bridge" by Fergie
"Single Ladies (Put A Ring On It)" by Beyonce
"Loba" by Shakira

Song suggestions generated with Last.fm Dataset, the song similarity dataset of the Million Song Dataset.

A Project by Randall Harris and Samantha Trippy

Northwestern University - Winter 2015
Professor Bryan Pardo - EECS 352: Machine Perception of Music and Audio
Contact Info:
Randall Harris
Samantha Trippy