Update README.md
Browse files
README.md
CHANGED
@@ -15,16 +15,14 @@ widget:
|
|
15 |
## This model is trained for GoEmotions dataset which contains labeled 58k Reddit comments with 28 emotions
|
16 |
- admiration, amusement, anger, annoyance, approval, caring, confusion, curiosity, desire, disappointment, disapproval, disgust, embarrassment, excitement, fear, gratitude, grief, joy, love, nervousness, optimism, pride, realization, relief, remorse, sadness, surprise + neutral
|
17 |
|
18 |
-
|
19 |
-
|
20 |
- The model works well on most of the emotions except: 'desire', 'disgust', 'embarrassment', 'excitement', 'fear', 'grief', 'nervousness', 'pride', 'relief', 'remorse', 'surprise']
|
21 |
- I'll try to fine-tune the model further and update here if RoBERTa achieves a better performance.
|
22 |
-
|
23 |
-
## Some Training details:
|
24 |
- Each text datapoint can have more than 1 label. Most of the training set had 1 label: Counter({1: 36308, 2: 6541, 3: 532, 4: 28, 5: 1})
|
25 |
-
|
26 |
|
27 |
-
##
|
28 |
============================================================<br>
|
29 |
Emotion: admiration<br>
|
30 |
============================================================<br>
|
|
|
15 |
## This model is trained for GoEmotions dataset which contains labeled 58k Reddit comments with 28 emotions
|
16 |
- admiration, amusement, anger, annoyance, approval, caring, confusion, curiosity, desire, disappointment, disapproval, disgust, embarrassment, excitement, fear, gratitude, grief, joy, love, nervousness, optimism, pride, realization, relief, remorse, sadness, surprise + neutral
|
17 |
|
18 |
+
## Training details:
|
19 |
+
- The training script is provided here: https://github.com/bsinghpratap/roberta_train_goEmotion
|
20 |
- The model works well on most of the emotions except: 'desire', 'disgust', 'embarrassment', 'excitement', 'fear', 'grief', 'nervousness', 'pride', 'relief', 'remorse', 'surprise']
|
21 |
- I'll try to fine-tune the model further and update here if RoBERTa achieves a better performance.
|
|
|
|
|
22 |
- Each text datapoint can have more than 1 label. Most of the training set had 1 label: Counter({1: 36308, 2: 6541, 3: 532, 4: 28, 5: 1})
|
23 |
+
-- So currently I just used the first label for each of the datapoint. Not ideal but it does a decent job.
|
24 |
|
25 |
+
## Model Performance
|
26 |
============================================================<br>
|
27 |
Emotion: admiration<br>
|
28 |
============================================================<br>
|