Emotion detection in Twitter Dataset

Supervised classification task is to detect emotions in raw text. We've downloaded and prepared data from two different sources.

Model architecture

Model inspired from here: https://arxiv.org/abs/1610.08815
This model is used on both datasets

Dataset I

Source: http://saifmohammad.com/WebPages/EmotionIntensity-SharedTask.html

	fear	joy	anger	sadness
Train	455	333	327	317
CV	114	83	82	79
Test	471	356	348	327
Total	569	416	409	396

Results

Classification report:

Metric Plot:

Loss Plot:

Conclusion

We've achieved an F1-score of 0.70 on test dataset

Dataset II

Source: https://www.kaggle.com/eray1yildiz/emotion-classification

	joy	sadness	anger	fear	love	surprise
Train	90282	77559	36683	30536	22114	9583
CV	22571	19390	9171	7634	5529	2395
Test	28214	24238	11463	9542	6911	2994
Total	141067	121187	57317	47712	34554	14972

Results

Classification report:

Metric Plot:

Loss Plot:

Conclusion

We've achieved an F1-score of 0.915 on test dataset