text records in the corpus
One dataset. Many ways to train with it.
A multi-version text emotion dataset prepared for classification, preprocessing experiments, baseline benchmarking, and NLP prototyping.
records represented across each complete preprocessing version
emotion categories
complete preprocessing variants
TF-IDF vocabulary size
Six signals. One text corpus.
The dataset separates text into six emotion categories while preserving enough volume for experimentation across multiple preprocessing strategies.
Anger
samples BOTH release
Disgust
samples BOTH release
Fear
represented in corpus
Happiness
represented in corpus
Sadness
represented in corpus
Surprise
represented in corpus
Same corpus. Different preprocessing.
Each release preserves the same classification goal while changing how the text is normalized before modeling.
Clean
Cleaned text with selected noise removed while preserving original casing and punctuation structure.
emotion_clean.csv
Lowercase
A normalized version where the text has been converted to lowercase for case-insensitive experiments.
emotion_lower.csv
No Punctuation
Punctuation is removed while casing is preserved, allowing isolated preprocessing comparisons.
emotion_nopunct.csv
Lower + No Punctuation
The most normalized current version, combining lowercase conversion with punctuation removal.
emotion_both.csv
See the prediction workflow in action.
Enter a sentence to explore how the EmotionText classification interface presents an emotion result. This portfolio deployment demonstrates the interface only.
The production classifier runs through a Python API. Live model inference is not connected to this static Netlify deployment.
Enter a sentence and run the interface preview to see how an emotion label would be presented.
INTERFACE DEMONSTRATION ONLY. THE LABEL SHOWN HERE IS GENERATED BY SIMPLE FRONTEND DEMO LOGIC, NOT BY THE TRAINED MACHINE LEARNING MODEL.
A baseline before deployment.
The dataset was evaluated through a TF-IDF-based multiclass classification experiment using Logistic Regression.
test accuracy
Reported from the current local training experiment using the prepared dataset.
High accuracy needs careful interpretation.
The dataset contains augmented and closely related text variants, which may increase similarity between training and testing examples. This result should therefore be treated as an internal experiment, not as a guarantee of performance on unseen real-world text.
Built for experimentation.
EmotionText is structured as a reusable NLP dataset rather than a single-model project.
Emotion Modeling
Train and compare multiclass classifiers across six emotion labels.
Text Normalization
Compare how casing, punctuation and cleaning strategies affect downstream models.
NLP Experiments
Use the prepared versions as a starting point for feature engineering, benchmarking and rapid NLP prototypes.
One corpus. Four preprocessing views.
Built as a structured dataset project for machine learning experiments, text preprocessing comparisons, and emotion-classification prototypes.