Tweet-Emotion-Classifier

Emotion Classifier — AI Got Talent

This project was built for Developers Day ‘26 hosted by FAST NUCES Karachi. The competition was called AI Got Talent and the challenge was simple — build a machine learning model that can detect emotions in tweets. We had 2 hours. No pretrained models. Just raw Python and a lot of coffee ☕

What does it do?

You give it a tweet. It tells you what emotions are in it.

Not just one emotion — a tweet can have multiple emotions at the same time. For example:

“I’m so proud of what we achieved, this feels amazing!” → pride, joy, admiration

“This situation is frustrating and honestly makes me really upset.” → anger, disgust, sadness

The model detects 9 emotions: admiration anger disgust fear hope joy love pride sadness

How we built it

We kept things simple but smart.

Text Cleaning First we clean the tweet — lowercase everything, remove links, mentions, hashtags, and fix weird characters. We also expand contractions like “I’m” → “I am” so the model understands better.

Feature Extraction We used TF-IDF vectorization in three ways:

Word level (unigrams, bigrams, trigrams)
Character level (catches spelling variations)
Bigram focused (phrase patterns)

All three are combined into one big feature matrix.

The Models We didn’t rely on just one model. We trained four:

Logistic Regression (C=5)
Logistic Regression (C=15)
LinearSVC
SGD Classifier

Then we combined their predictions using a weighted ensemble — basically letting all four vote together for better accuracy.

Threshold Tuning Instead of using a fixed 0.5 cutoff for every emotion, we tuned a separate threshold for each emotion. Some emotions like pride needed a lower threshold (0.25) while others like fear needed higher (0.50). This alone boosted our F1 score significantly.

Results

Metric	Score
F1 Micro	0.893
F1 Macro	0.893
Precision	0.886
Recall	0.901
Hamming Loss	0.049
Exact Match	66.7%

Per emotion breakdown:

Emotion	Precision	Recall	F1
admiration	0.943	0.983	0.962
anger	0.879	0.860	0.870
disgust	0.962	0.927	0.944
fear	0.944	0.919	0.932
hope	0.962	0.856	0.906
joy	0.759	0.837	0.796
love	0.871	0.908	0.889
pride	0.941	0.969	0.955
sadness	0.738	0.845	0.788

How to run it yourself

1. Clone the repo ```bash