Kaggle Dataset from: https://www.kaggle.com/praveengovi/emotions-dataset-for-nlp
The aim of this project is to determine the emotion that is associated with a given body of text. The emotions are the following:
- anger
- fear
- joy
- love
- sadness
- surprise
The plan of the analysis was to examine the accuracy results using the following vectorizers and to determine which produces the best accuracy:
- CountVectorizer: Emotions Prediction - CountVect.ipynb
- Term Frequency (TF) vectorizer (
TfidfVectorizer(use_idf=False)): Emotions Prediction - Tf.ipynb - TfidfVectorizer: Emotions Prediction - Tfidf.ipynb
| Vectorizer | Test Prediction Accuracy |
|---|---|
| CountVectorizer | 0.9654 |
| TF Vectorizer | 0.962 |
| TFIDF Vectorizer | 0.9256 |
The prediction accuracy, using all vectorizers, was impressive. It appears, however, that using the Inverse Document Frequency (idf) results in a slightly worse prediction accuracy. This reduced performance might be attributable to the fact that stop words and punctuation were removed before vectorization.
The analysis using the TfidfVectorizer bears repeating, only without the text cleaning and lemmatization to determine whether its performance would be better without the preprocessing of text.