Update README.md

README.md

@@ -1,75 +1,75 @@
-# 🎯 Tone Detection using `facebook/bart-large-mnli` (Zero-Shot Classification)
+# Zero-Shot Text Classification using `facebook/bart-large-mnli`
 
-This project demonstrates how to perform **Tone Detection** using the [`facebook/bart-large-mnli`](https://huggingface.co/facebook/bart-large-mnli) model through **zero-shot classification** based on Natural Language Inference (NLI).
+This repository demonstrates how to use the [`facebook/bart-large-mnli`](https://huggingface.co/facebook/bart-large-mnli) model for **zero-shot text classification** based on **natural language inference (NLI)**.
 
-This approach enables you to classify emotional tone (e.g., joy, anger, sadness) **without training**, by framing it as a textual entailment task.
+We extend the base usage by:
+- Using a labeled dataset for benchmarking
+- Performing optional fine-tuning
+- Quantizing the model to FP16
+- Scoring model performance
 
 ---
 
-## 📌 Model Details
+## 📌 Model Description
 
 - **Model:** `facebook/bart-large-mnli`
-- **Task:** Zero-shot classification via NLI
-- **Approach:** Checks if the input sentence entails a hypothesis (e.g., "This text expresses anger.")
-- **Strength:** No labeled training data required
+- **Type:** NLI-based zero-shot classifier
+- **Architecture:** BART (Bidirectional and Auto-Regressive Transformers)
+- **Usage:** Classifies text by scoring label hypotheses as NLI entailment
 
 ---
 
-## 📂 Dataset Used
+## 📂 Dataset
 
-For benchmarking and scoring, we use the [`go_emotions`](https://huggingface.co/datasets/go_emotions) dataset:
+We use the [`yahoo_answers_topics`](https://huggingface.co/datasets/yahoo_answers_topics) dataset from Hugging Face for evaluation. It contains questions categorized into 10 topics.
 
 ```python
 from datasets import load_dataset
 
-dataset = load_dataset("go_emotions")
+dataset = load_dataset("yahoo_answers_topics")
 ```
 
-# 🧠 Tone Detection (Inference)
-```Python
+# 🧠 Zero-Shot Classification Logic
+The model checks whether a text entails a hypothesis like:
+
+"This text is about sports."
+
+For each candidate label (e.g., "sports", "education", "health"), we convert them into such hypotheses and use the model to score them.
+
+# ✅ Example: Inference with Zero-Shot Pipeline
+```python
 from transformers import pipeline
 
 classifier = pipeline("zero-shot-classification", model="facebook/bart-large-mnli")
 
-labels = ["joy", "anger", "sadness", "fear", "surprise", "neutral"]
-
-text = "I can't believe this is happening again. So frustrating."
+sequence = "The team played well and won the championship."
+labels = ["sports", "politics", "education", "technology"]
 
-result = classifier(text, candidate_labels=labels, hypothesis_template="This text expresses {}.")
+result = classifier(sequence, candidate_labels=labels)
 print(result)
 ```
 
-# 🧪 Evaluation with Scoring
+# 📊 Scoring / Evaluation
+Evaluate zero-shot classification using accuracy or top-k accuracy:
 
 ```python
 from sklearn.metrics import accuracy_score
 
-# Mapping GoEmotions label indices to names
-id2label = dataset["train"].features["labels"].feature.names
-
-# Evaluate on a small sample
-def evaluate(dataset, candidate_labels):
+def evaluate_zero_shot(dataset, labels):
     correct = 0
     total = 0
-    for row in dataset.select(range(100)):  # Use more samples as needed
-        text = row["text"]
-        true_labels = [id2label[i] for i in row["labels"]]
-        result = classifier(text, candidate_labels=candidate_labels, hypothesis_template="This text expresses {}.")
+    for example in dataset:
+        result = classifier(example["question_content"], candidate_labels=labels)
         predicted = result["labels"][0]
-        if predicted in true_labels:
-            correct += 1
+        true = labels[example["topic"]]
+        correct += int(predicted == true)
         total += 1
-    return correct/total
-
-accuracy = evaluate(dataset["test"], candidate_labels=labels)
-print(f"Zero-shot Accuracy: {accuracy:.2%}")
-```
-
-# ⚙️ Use Cases
-Customer support tone analysis
-
-Chat moderation for emotional tone
+    return correct / total
 
-Feedback sentiment detection
+labels = ["Society & Culture", "Science & Mathematics", "Health", "Education", 
+          "Computers & Internet", "Sports", "Business & Finance", "Entertainment & Music", 
+          "Family & Relationships", "Politics & Government"]
 
-Real-time conversation emotion tagging
+acc = evaluate_zero_shot(dataset["test"].select(range(100)), labels)
+print(f"Accuracy: {acc:.2%}")
+```