EmuPlayer Music Recommendation System Based on User Emotion
Download
Report
Transcript EmuPlayer Music Recommendation System Based on User Emotion
EmuPlayer
Music Recommendation System
Based on User Emotion
Using Vital-sensor
KMSF- sunny
親:namachanさん
1
Motivation
Music – Emotion: mutual relationship
Users choose songs based on current
feelings
Playlist constantly expanding
difficulty in picking appropriate song
2
Requirement
Track User Emotion
Recommend by Sorting playlist based on
user’s current emotion
Sort songs by 2 factors
Relevancy to User Preference
Effect on User Emotion
3
Related research
Matching Music Mood
and User Emotion
(-) Emotion: declared by
user
(-) Subjectiveness on
Music Mood?
(-) Music Mood and User
Emotion are always the
same?
EmuPlayer
Emotion: automatically
detected
User Preference under
each emotion is studied
Recommendation
implying emotional
effect’s feedback
4
EmuPlayer: Example from User A
Before
NoRPulse = 76.73, NoRTemp = 33.6
Pulse = 70.04, Temp = 34.0 relax/~sleepy
Song Information
SongNo = 3
SongID = 4, Title = “So Close”
After
“Like”
Pulse = 79.0,
Score = 1
Temp = 34.39 pleasure
5
Approach
Vital
Information
Emo
Detector
User
Emotion
Recommender
Sorted
Playlist
EmuPlayer
Emotion Recognition
Music Recommendation
6
Emotion Recognition
Merit of Vital-sensor
Requirements
Portability if integrated in Music Player
Continuity of output data
Sensitiveness towards changes in emotion
Vital-sensor meets all those requirement
7
Emotion Recognition
Vital-sensor v.s Other Methods
Portability
if integrated in
Music Player
Continuity
of output data
Sensitiveness
towards
changes in
emotion while
listening to
music
Facial Expression
X
X
△
Speech
△
X
△
Eyes movement
△
△
△
Brainwave
X
○
◎
Gesture
X
X
△
Vital-sensor
○
○
○
Requirement
regard for the use
in MRS
Method
8
Emotion Recognition
Russell model and Two Biosignals
Russell’s model
Horizontal axis:
Pleasure
SkinTemp
Vertical axis:
Arousal
Heart Rate
90°
135°
45°
180°
0°
225°
315°
270°
9
Emotion Recognition
Mapping in EmuPlayer
Define Emotion Region
Based on Theory of the
Fuzziness of words
8 equal regions
Mapping
Based on angle
(1,1)45°Excitement
10
Music Recommendation
2 factors to evaluate a song
Relevancy to User Preference
Mental Effect on User Emotion
2 subjects
Study User Preference
Study Emotional Effect of songs
11
Music Recommendation
Study User Preference
Rating Like/Dislike
Record listening history
12
Music Recommendation
Study songs’ emotional effect
Define emotional
effect: Good-Bad
to avoid potentially
harmful
recommendations to
user emotion
good
bad
13
Music Recommendation
Effect Definition Survey
Matching point = 42/48*100 = 87.5%
14
Music Recommendation
Rating songs
Better songs rank higher
15
System Flow
2
Data
Pre
Processor
Data Receiver
8
3
4
Emo
Detector
9
Recommender
10
7’
11
1
Database
5
5
10’
7
RF-ECG sensor
Evaluator
6
Interface
User
16
Demonstration
17
Evaluation on Emotion Recognition
Number of
Participant
Gender
Average Age
10~
Male
21
Experiment 1: Testing Accuracy of Emotion
Recognition through arranged situation
Survey: (1) if they experienced the emotion expressed
through the situation,
and (2) if not, what emotions rather than the one in (1)
they experienced.
18
Result of Experiment 1
Output
(Result from
Engine)
Arousal
Exciteme
nt
Pleasure
Relaxatio
n
Sleepine
ss
Disples
ure
Distress
Depressi
on
Input
(Verified
Experimenting Emo)
Relaxation
4.25% 7.96% 22.77
%
55.58
%
9.44% 0
0
0
Excitement
10%
63.34
%
21.66
%
5%
0
0
0
0
Pleasure
1.4%
10.87
%
66.86
%
20.87
%
0
0
0
0
Arousal
81.68
%
5%
0
6.66% 6.66% 0
0
0
Depression
8.33% 0
0
0
10%
55.01
%
Accuracy = 64.5%
6.66% 20%
19
Evaluation on Emotion Recognition
Experiment 2
Change User Emotion by music
Purpose
Verify whether the system can realize user’s
emotional changes
Verify songs’ influence on listeners’ emotions
20
Case
Experiment
Mean
Result
1
Arousal
Pleasure/Rela
xation
Classical
music
Arousal:81.68%1.41%
Pleasure:0%54.91%
Relaxation:6.66%38.94%
2
Normal
Pleasure
Music
participants
like
Pleasure: 66.86% 93.33%
3
Normal
Excitement
Fast beat
Music
Excitement: 10.87%62.12%
4
Normal
Depression
Loud Heavy Pleasure:66.86%Depression:
Music played 80.02%
in long time
21
Emotion Recognition
Conclusion from 2 experiments
Accuracy of Extracting Emotion: 64.5%
Strong at detecting bad emotions
Detect precisely regarding to changes of
user emotion
Hypothesis of music influencing on user
emotion is true
22
Evaluation: EmuPlayer Performance
Observing high-rating songs
% being “dislike” after being listened
% paying “bad influence” on user’s emotion
after being listened
% being reduced in score after being listened
Observing “like” song
Emotional change
23
EmuPlayer Performance
Observing high-rating songs
24
EmuPlayer Performance
Observing “like” song
Emo Before
Emo After
Percentage
Arousal
Relaxation
1%
Arousal
Sleepiness
1%
Excitement
Arousal
1%
Excitement
Pleasure
2%
Pleasure
Excitement
2%
Pleasure
Pleasure
77%
Relaxation
Arousal
1%
Relaxation
Pleasure
2%
Relaxation
Relaxation
7%
Sleepiness
Sleepiness
4%
Displeasure
Displeasure
2%
No song influencing badly on users’ emotion
25
Conclusion of EmuPlayer Performance
EmuPlayer algorithm ensures
recommendation of songs meeting
proposed two requirements
Songs influencing badly on user emotion: 0%
Songs being “dislike” in later listening time:
6.66%
26
Overall survey
Are you interested in such a MRS system?
Yes: 90%
Scale your satisfaction of EmuPlayer’s work
Average point = 3.6/5
Do you feel uncomfortable wearing RF-ECG?
Yes: 40%
Did you experience bad emotion after listening to highrating song
Yes: 10%
Reflect the truth: proposing ER method responds to only clear
and strong emotions. Slight changes in emotion felt by users
may not be recognized by the system
27
Conclusion
Concept of EmuPlayer is essential
Evaluate song through 2 factors
Employ User Emotion as crucial input for MRS
Accuracy of extracting emotion is not very high:
64.5%
Strong at detecting bad emotion
Applicable in giving alert when playing music influences
badly on listener’s emotional state
EmuPlayer’s efficiency in suggesting songs
meeting the two requirements
28
Future works
Enhance the accuracy of detecting emotion by
Employing other means than Heart Rate and Skin
Temperature
Alternate RF-ECG
Enhance the work of Recommending music by
combining proposed method with songs’ content
analyzing
Enhance reasoning user’s state by combining
User Emotion with context analyzing
29
Thank you for listening
30