Can relevant

MUSIC
be delivered based on

PERSONALITY
of a person ?
 Personality Based Music
Recommendation System
Abhishek Paudel (070BCT502)
Brihat Ratna Bajracharya (070BCT513)
Miran Ghimire (070BCT521)
Nabin Bhattarai (070BCT522)
Into how many categories can
PERSONALITY
be classified ?

Personality Based Music Recommendation System 3

 4000
Gorden Allport

Personality Based Music Recommendation System 4

 4000

16
Raymond Cattell

Personality Based Music Recommendation System 5

 4000

16

3
Hans Eysenck

Personality Based Music Recommendation System 6

 4000

16

5
Core Personality Traits

Personality Based Music Recommendation System 7

 Big 5 Personality Traits

Openness to Conscientiousness Extroversion Agreeableness Neuroticism

Experience
inventive/curious efficient/organized outgoing/energetic friendly/compassionate sensitive/nervous

Personality Based Music Recommendation System 8

 Classification
is based on user’s
Facebook status updates.

Personality Based Music Recommendation System 9

Data Preprocessing
Facebook
Status Updates

Stop words
Tokenization Stemming
removal

Feature
Feature PoS tagging
Vector
Vector & word
Creation filtering
Personality Based Music Recommendation System 10
Personality Classification

Openness

Conscientiousness
Naïve Bayes/
Feature Logistic
Vector Regression Extroversion
Classifiers
Agreeableness
Neuroticism

Personality Based Music Recommendation System 11

Evaluation of Classifiers
Naïve Predicted Logistic Predicted
Bayes Regression
Classifier Yes No Classifier Yes No

Yes 64 59 Yes 57 66
Actual

Actual
No 52 75 No 61 66

F-measure: 0.5356 F-measure: 0.4730

Accuracy: 0.5560 Accuracy: 0.4920

Naïve Bayes Classifier performed better that Logistic Regression Classifier.

Personality Based Music Recommendation System 12
Personality  Music Recommendation
using Collaborative Filtering

Personality Based Music Recommendation System 13

Music Recommendation

Users Music
User 1 User 2 Music 1 Music 2
User 3 ...…… Music 3 ...……

Get Similar
Get Ranking
Users based
Estimated of Music Recommend
on
User Personality
Rating based based on
on Rating by estimated Top N Music
Scores/Music
Similar Users Rating
Ratings

Personality Based Music Recommendation System 14

Recommendation Models
CF with
Rating
Matrix
CF with
Global Average
CF with Baseline + Global
Average CF with Baseline
Average CF with
Average
CF with
Personality
Matrix

Matrix
Factorization
Personality Based Music Recommendation System 15
Comparison of RMSE of different
Recommendation Model in the system
5
4.5
4
3.5
3
2.5
2
1.5
1
0.5
0
User to User
Collaborative
User to User Filtering with
User to User User to User
Collaborative Weighted
User to User User to User Collaborative Collaborative
Filtering with Average of User
Collaborative Collaborative Filtering with Filtering with
Global Baseline Weighted Personality Matrix
Filtering with Filtering with User Rating User Personality
Algorithm Average of User Matrix and User Factorization
User Rating User Personality Matrix with Matrix with
Personality rating Matrix
Matrix Matrix combination of combination of
Matrix and User with
Global Baseline Global Baseline
Rating Matrix combination of
Global Baseline
Algorithm
RMSE 2.86 3.89 3.2 4.72 3.1 3.2 3.04 0.88
Personality Based Music Recommendation System 16
Findings
▪ Personality had the positive impact upon the user-
user collaborative filtering where similar user are
computed with the user rating matrix.
▪ However, it still lags behind the matrix factorization
model in terms of RMSE.

Personality Based Music Recommendation System 17

Limitations
▪The problem that still prevails, in the user to
user collaborative filtering with personality
are:

▪ Cold start problem in the item

▪ Gray sheep problem

Personality Based Music Recommendation System 18

Future Enhancements
▪Classifier can be improved by
▪ Considering emojis ☺
▪ Considering users’ likes, interests and activities

▪Recommendation can be improved by

▪ Content based Filtering for solving Cold Start
problem for item

Personality Based Music Recommendation System 19

Conclusion
▪The personality traits of the user could be one
of the crucial factor in computation of similar
user in user to user collaborative filtering
where similar user are computed only via user
rating matrix.

Personality Based Music Recommendation System 20