YatishW79/classical-ml-goemotions

Classical ML — GoEmotions Emotion Classifier

TF-IDF + OneVsRest classifiers for multi-label emotion classification on GoEmotions (28 labels).

CS6120 NLP Project — Northeastern University

Models

Model	Micro-F1	Macro-F1	Subset Acc	Hamming Acc
Logistic Regression	0.4448	0.2388	0.3156	0.9658
LinearSVC (calibrated)	0.5233	0.4266	0.3097	0.9542
XGBoost	0.5485	0.4044	0.4133	0.9622

Files

• tfidf_vectorizer.pkl — TF-IDF vectorizer (15k features, bigrams, sublinear TF)
• mlb.pkl — MultiLabelBinarizer (28 GoEmotions labels)
• logistic_regression.pkl — OneVsRest Logistic Regression
• linearsvc.pkl — OneVsRest LinearSVC (Platt-calibrated)
• xgboost.pkl — OneVsRest XGBoost
• thresholds_lr.pkl — per-class optimal thresholds for LR (numpy array, shape 28)
• thresholds_linearsvc.pkl — per-class optimal thresholds for LinearSVC (numpy array, shape 28)

Usage

import joblib, numpy as np

tfidf      = joblib.load("tfidf_vectorizer.pkl")
mlb        = joblib.load("mlb.pkl")
model      = joblib.load("xgboost.pkl")        # or logistic_regression / linearsvc
thresholds = joblib.load("thresholds_lr.pkl")  # for LR; skip for XGBoost (use 0.30)

X = tfidf.transform(["I am so happy and grateful today!"])
proba  = model.predict_proba(X)
preds  = (proba >= thresholds).astype(int)     # shape (1, 28)
labels = mlb.classes_[preds[0].astype(bool)]
print(labels)