Dataset - Here

For this repository we have used Anger and Joy classes.

All dataset can be downloaded from the above website.

-----------------------------------Approachs------------------------------------------

Two approachs.

1. Based on dataset.
   a. First used Raw data.
   b. Pre-processed all dataset.(Explained Below)

2. Based on Vocabulary
   a. Used simple Stopword to create TfidfVectorizer.
   b. Used cutome Vocabulary created using unigrams and bigrams from tweet data.
   c. Used Twitter Glove Embedding.

-----------------------------------Models--------------------------------------------

Three model

1. Random Forest Regressor.
2. SVM.
3. Simple MLP using keras(run for 100 epochs).

All code implemention is given in Jupyter notebook with relevent comments.

-----------------------------------Results-------------------------------------------

Following are the best result obtain in test set for each approach.

For Anger Dataset

1. Without Pre-processing.
   a. simple Stopword.

       Best result - Random Forest
               metrics     Value
       0          Pearsonr  0.247100
       1         Spearmanr  0.222550
       2   Pearsonr >= 0.5  0.136760
       3  Spearmanr >= 0.5  0.109948    
   

   b. cutome Vocabulary.

       Best result - Random Forest
                metrics     Value
       0          Pearsonr  0.536161
       1         Spearmanr  0.504543
       2   Pearsonr >= 0.5  0.400112
       3  Spearmanr >= 0.5  0.363594
   

   c. Twitter Glove Embedding.

       Best result - SVM
             metrics     Value
       0          Pearsonr  0.586624
       1         Spearmanr  0.580586
       2   Pearsonr >= 0.5  0.429999
       3  Spearmanr >= 0.5  0.424292
   

2. With Pre-processing a. simple Stopword.

       Best result - Random Forest
                      metrics     Value
       0          Pearsonr  0.243917
       1         Spearmanr  0.217084
       2   Pearsonr >= 0.5  0.128100
       3  Spearmanr >= 0.5  0.110319
      ```
   b. cutome Vocabulary.
   

    Best result - Random Forest
                metrics     Value
    0          Pearsonr  0.535899
    1         Spearmanr  0.504324
    2   Pearsonr >= 0.5  0.399039
    3  Spearmanr >= 0.5  0.360943
    ```
   

   c. Twitter Glove Embedding.

       Best result - SVM
                   metrics     Value
       0          Pearsonr  0.587715
       1         Spearmanr  0.581532
       2   Pearsonr >= 0.5  0.432472
       3  Spearmanr >= 0.5  0.426328
       ```
       
   

For Joy Dataset

1. Without Pre-processing. a. simple Stopword.

       Best result - Random Forest
                   metrics     Value
       0          Pearsonr  0.312417
       1         Spearmanr  0.309787
       2   Pearsonr >= 0.5  0.154492
       3  Spearmanr >= 0.5  0.160097
       ```
   b. cutome Vocabulary.
   

    Best result - SVM
                metrics     Value
    0          Pearsonr  0.474524
    1         Spearmanr  0.494809
    2   Pearsonr >= 0.5  0.308389
    3  Spearmanr >= 0.5  0.307756
    ```
   

   c. Twitter Glove Embedding.

       Best result - SVM
               metrics     Value
       0          Pearsonr  0.606882
       1         Spearmanr  0.604149
       2   Pearsonr >= 0.5  0.449256
       3  Spearmanr >= 0.5  0.442449
       ```
   
   

2. With Pre-processing a. simple Stopword.

       Best result - Random Forest
           metrics     Value
       0          Pearsonr  0.257611
       1         Spearmanr  0.256136
       2   Pearsonr >= 0.5  0.157086
       3  Spearmanr >= 0.5  0.154409
       ```
   b. cutome Vocabulary.
   

    Best result - SVM
                metrics     Value
    0          Pearsonr  0.482738
    1         Spearmanr  0.499942
    2   Pearsonr >= 0.5  0.310332
    3  Spearmanr >= 0.5  0.323588
    ```
   

   c. Twitter Glove Embedding.

       Best result - SVM
                metrics     Value
       0          Pearsonr  0.579386
       1         Spearmanr  0.576765
       2   Pearsonr >= 0.5  0.382190
       3  Spearmanr >= 0.5  0.382472
       ```