Introducing
Prezi AI.
Your new presentation assistant.
Refine, enhance, and tailor your content, source relevant images, and edit visuals quicker than ever before.
Trending searches
Loading…
1
Transcript
Naive Bayes Classifier
for Music Emotion Classification
Based on Lyrics
CSE 437
Paper Review Presention
GROUP - 14
INTRODUCTION
Text Classification
- Music emotion classification - text classification
- Better Algorithm
- Naive Bayesian Classifier
Introduction
Naive Bayes
Text Classificaton
- Researched on chinese text categorization
- Assumed words of lyrics are equal weight and independent
- found conditional probability of each category
Naive Bayesian Classifier
- Based on bayes theorem with independence assumptions
The Naive Bayes classifier, which uses the Naive Bayesian
formula to calculate the probability of each class A given the values B_i of all attributes for an instance to be classified, the conditional independence of the attributes given the class
Web Crawling
Text segmentation
Methodology
Methodology
Train & Test
Naive Bayes Classifier
Accuracy
Data Collection
Dataset
- Lyrics Collection
- Emotion Model
- Mapping of emotion models
Lyrics Collection
point 1
- Baidu music emotion labels
- emotion labels - sad, passionate, quiet, comfortable, sweet, inspirational, lonely, miss, romantic, yearning, joyful, soulful, happy, nostalgic, relaxed
- Web crawling - Scrapy-a python framework
Emotion Model
- Two models defined by dimension
- Russell Emotion Model
- Thayer emotion Model
point 2
Russell Emotion Model
- Valence refers to the positive and negative degree of emotion
- Arousal refers to the intensity of emotion.
Model 1
Thayer Emotion Model
- Energy refers to the volume or intensity of sound in music
- Stress refers to the tonality and tempo of music.
Model 2
Mapping Emotion Model
point 3
- Transformed emotion labels
- contentment
- depression
- exuberance
- Removed ambiguous labels
- 3552 songs with lyrics & label
- contentment - 346
- depression - 2175
- exuberance - 1075
D-1
EXPERIMENTS
- Lyrics were segmented to pick up emotional words
- Jeiba module was used
- 4 different datasets
- 2369 for training
- 1183 for testing
- performance of model evaluated by their accuracy
Experiments
RESULT OF D-1
- Dataset with both english and chinese lyrics
RESULT OF D-2
Dataset with only chinese words
D-2
RESULT OF D-3
Dataset with both chinese and eglish lyrics
D-3
RESULT OF D-4
Dataset with only chinese lyrics
D-4
DATASETS ACCURACY
Accuracy
Conclusion
- Described principle and reason of using bayesian algorithm
- Focused on designing an effective classifier
- Expect to classify the music using audio
Conclusion
Choose a template
Presentations from around the world
Learn more about creating dynamic, engaging presentations with Prezi