Fine-tune on MYUniDialectSentiment840 (840 samples, 14 dialects). Test acc=0.9796, F1_macro=0.9753.

README.md

@@ -1,253 +1,61 @@
 ---
 language:
-- en
 - ms
-license: apache-2.0
+- en
 tags:
-- emotion-classification
-- multilingual
-- xlm-roberta
 - text-classification
-- sentiment-analysis
-- malay
-- english
-- production-ready
-- gradio-app
-- interactive-demo
-datasets:
-- custom-multilingual-emotion-dataset
-metrics:
-- accuracy
-- f1
-- precision
-- recall
+- emotion
+- malaysian
+- bahasa-malaysia
+- code-switching
+- educational
+license: apache-2.0
+pipeline_tag: text-classification
 ---
 
-# 🎭 Multilingual Emotion Classifier - Interactive Demo Available!
-
-## 🚀 **TRY THE INTERACTIVE APP NOW!**
-
-**🎮 [Launch Interactive Demo](https://huggingface.co/spaces/rmtariq/multilingual-emotion-classifier-demo)** ← Click to test the model in your browser!
-
-A state-of-the-art multilingual emotion classification model with **85.0% accuracy**, **comprehensive Malay language support**, and **interactive web interface**.
-
-## 🎭 **INTERACTIVE FEATURES**
-
-### **🎯 Single Text Analysis**
-- Real-time emotion classification
-- Confidence scoring with visual gauges
-- Support for English and Malay
-- Interactive charts and visualizations
-
-### **📊 Batch Processing**
-- Analyze multiple texts simultaneously
-- Emotion distribution charts
-- Detailed results tables
-- Export capabilities
-
-### **🧪 Model Testing**
-- Run predefined test cases
-- Validate model performance
-- Check accuracy across languages
-- Verify all fixes are working
-
-### **📱 User-Friendly Interface**
-- Beautiful, responsive design
-- No coding required
-- Instant results
-- Mobile-friendly
-
-## 🚀 **Quick Start**
-
-### **Option 1: Interactive Web App (Recommended)**
-🎮 **[Try the Interactive Demo](https://huggingface.co/spaces/rmtariq/multilingual-emotion-classifier-demo)**
-
-No installation required! Test the model directly in your browser.
-
-### **Option 2: Python Code**
-```python
-from transformers import pipeline
-
-classifier = pipeline(
-    "text-classification", 
-    model="rmtariq/multilingual-emotion-classifier"
-)
-
-# English examples
-result = classifier("I am so happy today!")
-print(result)  # [{'label': 'happy', 'score': 0.999}]
-
-# Malay examples (now working perfectly!)
-result = classifier("Ini adalah hari jadi terbaik!")
-print(result)  # [{'label': 'happy', 'score': 0.999}] ✅
-
-result = classifier("Terbaik!")
-print(result)  # [{'label': 'happy', 'score': 0.999}] ✅
-```
-
-## 📊 **Performance Highlights**
-- ✅ **Overall Accuracy**: 85.0%
-- ✅ **F1 Macro Score**: 85.5%
-- ✅ **English Performance**: 100.0% accuracy
-- ✅ **Malay Performance**: 100% (all issues fixed)
-- ✅ **Speed**: 20+ predictions/second
-- ✅ **Interactive Demo**: Available for instant testing
-
-## 🎯 **Supported Emotions**
-
-| Emotion | Emoji | English Example | Malay Example | Demo Result |
-|---------|-------|-----------------|---------------|-------------|
-| **anger** | 😠 | "I'm so angry!" | "Marah betul!" | Try in demo! |
-| **fear** | 😨 | "I'm scared!" | "Takut sangat!" | Try in demo! |
-| **happy** | 😊 | "I'm so happy!" | "Gembira sangat!" | Try in demo! |
-| **love** | ❤️ | "I love you!" | "Sayang kamu!" | Try in demo! |
-| **sadness** | 😢 | "I'm so sad" | "Sedih betul" | Try in demo! |
-| **surprise** | 😲 | "What a surprise!" | "Terkejut betul!" | Try in demo! |
+# rmtariq/multilingual-emotion-classifier
 
-## 🔧 **What Was Fixed (Version 2.1)**
+Fine-tuned emotion classifier (7-class) for Malaysian higher-education
+feedback. Trained on **MYUniDialectSentiment840 (840 samples, 14 dialects, 20 topics, 15 learning contexts)**, a hand-curated balanced corpus covering
+Standard Malay, 13 regional dialects, and Manglish code-switching.
 
-### **Test These in the Interactive Demo:**
-```python
-# Before Fix (Problematic) → After Fix (Perfect)
-"Ini adalah hari jadi terbaik"  # ❌ anger → ✅ happy (99.9%)
-"Terbaik!"                      # ❌ surprise → ✅ happy (99.9%)
-"Ini adalah hari yang baik"     # ❌ anger → ✅ happy (99.9%)
-```
+## Labels
+- `happy`
+- `love`
+- `surprise`
+- `no_clear_emotion`
+- `sadness`
+- `anger`
+- `fear`
 
-**🎮 [Test these fixes in the interactive demo!](https://huggingface.co/spaces/rmtariq/multilingual-emotion-classifier-demo)**
+## Held-out test metrics (n=147, stratified)
 
-## 🧪 **Comprehensive Testing Suite**
+| split | accuracy | f1_macro |
+|---|---:|---:|
+| validation (n=105) | 0.9524 | 0.9440 |
+| test (n=147) | 0.9796 | 0.9753 |
 
-### **Interactive Testing (New!)**
-🎮 **[Launch Interactive Demo](https://huggingface.co/spaces/rmtariq/multilingual-emotion-classifier-demo)**
-- Web-based testing interface
-- Real-time results
-- Visual confidence charts
-- Batch processing capabilities
+## Intended use
 
-### **Command Line Testing**
-```bash
-# Install requirements
-pip install torch transformers numpy pandas scikit-learn
+Sentiment / emotion monitoring of student feedback for Malaysian
+higher-education institutions. Designed to handle code-switched,
+dialect-heavy and informal academic discourse.
 
-# Quick test (30 seconds)
-python test_model.py --test-type quick
+## Training details
 
-# Comprehensive test (2 minutes)
-python test_model.py --test-type comprehensive
+- Base: previous revision of `rmtariq/multilingual-emotion-classifier`
+- Optimizer: AdamW (lr=2e-5, weight_decay=0.01, warmup_ratio=0.1)
+- Epochs: 5 with early stopping on validation macro-F1
+- Batch size: 16 (train) / 32 (eval), max_length=128
+- Hardware: Apple Silicon MPS
+- Class-weighted cross-entropy (for emotion only)
 
-# Interactive terminal mode
-python test_model.py --test-type interactive
-```
+## Dataset
 
-### **Automated Validation**
-```bash
-# Run automated validation
-python validate_model.py
-
-# Generate validation report
-python validate_model.py --output validation_report.json
-```
-
-## 🏭 **Production Use Cases**
-
-### **✅ Social Media Monitoring**
-```python
-# Real-time emotion analysis
-social_posts = [
-    "Love the new update! 😍",
-    "Suka sangat dengan produk ni!",
-    "This is frustrating...",
-    "Kecewa dengan service"
-]
-
-emotions = classifier(social_posts)
-# Or test in the interactive demo!
-```
-
-### **✅ Customer Service Automation**
-```python
-# Automated ticket routing
-support_messages = [
-    "I'm really upset about this issue",
-    "Marah betul dengan masalah ni",
-    "Thank you for the great service!",
-    "Terima kasih, service terbaik!"
-]
-
-# Route high-emotion tickets to human agents
-# Test these examples in the interactive demo!
-```
-
-## 📈 **Performance Evolution**
-
-| Phase | Accuracy | F1 Macro | Interactive Demo |
-|-------|----------|----------|------------------|
-| **Initial Baseline** | 17.5% | 8.7% | ❌ Not available |
-| **Phase 1 Optimization** | 68.7% | 34.0% | ❌ Not available |
-| **Phase 2 Optimized** | **85.0%** | **85.5%** | ❌ Not available |
-| **Phase 3 + Interactive** | **85.0%** | **85.5%** | ✅ **Available!** |
-
-**Total Improvement**: **4.9x performance gain** + **interactive web interface**
-
-## 🎮 **How to Use the Interactive Demo**
-
-1. **🔗 [Click here to open the demo](https://huggingface.co/spaces/rmtariq/multilingual-emotion-classifier-demo)**
-2. **Choose a tab**: Single Text, Batch Analysis, or Model Testing
-3. **Enter your text**: Type in English or Malay
-4. **Get instant results**: See emotions, confidence, and visualizations
-5. **Explore features**: Try batch processing and predefined tests
-
-## 🧪 **Demo Features**
-
-### **🎯 Single Text Analysis**
-- Enter any text in English or Malay
-- Get instant emotion classification
-- See confidence scores with visual gauges
-- Try the provided examples
-
-### **📊 Batch Analysis**
-- Process multiple texts at once
-- See emotion distribution charts
-- Get detailed results tables
-- Perfect for analyzing conversations or reviews
-
-### **🧪 Model Testing**
-- Run 13 predefined test cases
-- Validate model performance
-- Check that all fixes are working
-- See accuracy metrics
-
-## 📞 **Contact & Resources**
-
-- **🎮 Interactive Demo**: [Launch App](https://huggingface.co/spaces/rmtariq/multilingual-emotion-classifier-demo)
-- **📖 Testing Guide**: See `TESTING_GUIDE.md` for comprehensive testing
-- **🤖 Model Repository**: [rmtariq/multilingual-emotion-classifier](https://huggingface.co/rmtariq/multilingual-emotion-classifier)
-- **👨‍💻 Author**: rmtariq
-
-## 📚 **Citation**
-
-```bibtex
-@misc{rmtariq2024multilingual_interactive,
-  title={Interactive Multilingual Emotion Classification with Web Demo},
-  author={rmtariq},
-  year={2024},
-  publisher={Hugging Face},
-  url={https://huggingface.co/rmtariq/multilingual-emotion-classifier},
-  note={Version 2.1 with interactive Gradio demo}
-}
-```
-
-## 📄 **License**
-
-This model is released under the Apache 2.0 License.
-
----
+`MYUniDialectSentiment840` — 840 samples, balanced on sentiment, stratified
+70/12.5/17.5 train/val/test by sentiment-x-dialect.
 
-**🎯 Status**: Production Ready with Interactive Demo ✅  
-**🚀 Performance**: 85.0% Accuracy, 85.5% F1 Macro  
-**🌍 Languages**: English, Malay (Fully Fixed)  
-**🎮 Demo**: Interactive Web Interface Available  
-**📅 Last Updated**: June 2024 (Version 2.1)  
+## Citation / authors
 
-**🎭 [Try the Interactive Demo Now!](https://huggingface.co/spaces/rmtariq/multilingual-emotion-classifier-demo)**
+- Raja Mohd Tariqi B. Raja Lope Ahmad — Ts., Fiscal Digest Sdn. Bhd.
+- Raja Qatrun Nada Bin Raja Mohd Tariqi — Master of Education, UKM

config.json

@@ -5,27 +5,30 @@
   "attention_probs_dropout_prob": 0.1,
   "bos_token_id": 0,
   "classifier_dropout": null,
+  "dtype": "float32",
   "eos_token_id": 2,
   "hidden_act": "gelu",
   "hidden_dropout_prob": 0.1,
   "hidden_size": 768,
   "id2label": {
-    "0": "anger",
-    "1": "fear",
-    "2": "happy",
-    "3": "love",
+    "0": "happy",
+    "1": "love",
+    "2": "surprise",
+    "3": "no_clear_emotion",
     "4": "sadness",
-    "5": "surprise"
+    "5": "anger",
+    "6": "fear"
   },
   "initializer_range": 0.02,
   "intermediate_size": 3072,
   "label2id": {
-    "anger": 0,
-    "fear": 1,
-    "happy": 2,
-    "love": 3,
+    "anger": 5,
+    "fear": 6,
+    "happy": 0,
+    "love": 1,
+    "no_clear_emotion": 3,
     "sadness": 4,
-    "surprise": 5
+    "surprise": 2
   },
   "layer_norm_eps": 1e-05,
   "max_position_embeddings": 514,
@@ -36,8 +39,7 @@
   "pad_token_id": 1,
   "position_embedding_type": "absolute",
   "problem_type": "single_label_classification",
-  "torch_dtype": "float32",
-  "transformers_version": "4.52.4",
+  "transformers_version": "4.57.6",
   "type_vocab_size": 1,
   "use_cache": true,
   "vocab_size": 250002

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8136644467afb9266338a23b2941df5c61dab14a43ab49356654a959f335b61d
-size 1112217312
+oid sha256:1d60839eba81934d24fdd0de8df4607b215e71954fe0133eefe2c876011c2c81
+size 1112220388

tokenizer.json

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:94b6f66edb763a1de9ee0edb887c3c33d2522e18a9fad2199600e485832af325
-size 17082987
+oid sha256:e63c7a5faf7f3257542d756311f64572a52913c2845c4a3849529e1ef87f8eef
+size 17082832

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f43ab5a58a6bdafe9a63f3a5ba9437d45634e88fdd715a699af80782bef28593
+size 5905

training_metrics.json

@@ -0,0 +1,40 @@
+{
+  "validation": {
+    "eval_loss": 0.2610369622707367,
+    "eval_accuracy": 0.9523809523809523,
+    "eval_f1_macro": 0.9440484327702372,
+    "eval_f1_weighted": 0.9526608729616249,
+    "eval_runtime": 0.1854,
+    "eval_samples_per_second": 566.462,
+    "eval_steps_per_second": 21.58,
+    "epoch": 5.0
+  },
+  "test": {
+    "test_loss": 0.16173337399959564,
+    "test_accuracy": 0.9795918367346939,
+    "test_f1_macro": 0.9753212396069539,
+    "test_f1_weighted": 0.9800003599323326,
+    "test_runtime": 0.3085,
+    "test_samples_per_second": 476.463,
+    "test_steps_per_second": 16.206,
+    "epoch": 5.0
+  },
+  "class_weights": [
+    0.8235294117647058,
+    1.2352941176470589,
+    2.1,
+    0.46153846153846156,
+    1.1506849315068493,
+    1.2173913043478262,
+    1.5555555555555556
+  ],
+  "label2id": {
+    "happy": 0,
+    "love": 1,
+    "surprise": 2,
+    "no_clear_emotion": 3,
+    "sadness": 4,
+    "anger": 5,
+    "fear": 6
+  }
+}