Updated by Author

README.md

@@ -210,6 +210,81 @@ The training utilized **Google Colab GPUs, which provided the necessary computat
 The training process was carried out using **PyTorch** as the primary framework, leveraging libraries such as **Hugging Face Transformers** for model implementation and training.
 
 
+## ROUGE Evaluation
+
+To evaluate the quality of the generated summaries, we employed the ROUGE (Recall-Oriented Understudy for Gisting Evaluation) scoring system. This method compares the generated summaries against reference summaries to quantify their similarity and overall quality.
+
+### Evaluation Code
+
+We used the `rouge_score` library to compute the ROUGE scores for our summaries. Below is the implementation:
+
+```python
+from rouge_score import rouge_scorer
+
+reference_summaries = [
+    "AI systems in healthcare improve diagnostics and personalize treatments.",
+    "Algorithms analyze market trends and help in fraud detection.",
+]
+
+generated_summaries = [
+    "In healthcare, AI systems are used for predictive analytics and improving diagnostics.",
+    "In finance, algorithms analyze market trends and assist in fraud detection."
+]
+
+scorer = rouge_scorer.RougeScorer(['rouge1', 'rouge2', 'rougeL'], use_stemmer=True)
+
+for reference, generated in zip(reference_summaries, generated_summaries):
+    scores = scorer.score(reference, generated)
+    print(f"Reference: {reference}")
+    print(f"Generated: {generated}")
+    print(f"ROUGE Scores: {scores}\n")
+```
+
+### ROUGE Scores
+
+#### Summary 1
+- **Reference**: "AI systems in healthcare improve diagnostics and personalize treatments."
+- **Generated**: "In healthcare, AI systems are used for predictive analytics and improving diagnostics."
+  
+**ROUGE-1**:
+- Precision: 72.73%
+- Recall: 88.89%
+- F1-Score: 80.00%
+
+This score indicates a strong overlap, showing that the generated summary captures a significant amount of relevant information.
+
+**ROUGE-2**:
+- Precision: 60.00%
+- Recall: 75.00%
+- F1-Score: 66.67%
+
+This indicates a good capture of bigrams, reflecting the generated summary's effectiveness in retaining key phrases.
+
+**ROUGE-L**:
+- Precision: 72.73%
+- Recall: 88.89%
+- F1-Score: 80.00%
+
+This score confirms that the sequence of words in the generated summary closely follows that of the reference.
+
+#### Summary 2
+- **Reference**: "Algorithms analyze market trends and help in fraud detection."
+- **Generated**: "In finance, algorithms analyze market trends and assist in fraud detection."
+  
+**ROUGE-1**:
+- Precision: 72.73%
+- Recall: 88.89%
+- F1-Score: 80.00%
+
+**ROUGE-2**:
+- Precision: 60.00%
+- Recall: 75.00%
+- F1-Score: 66.67%
+
+**ROUGE-L**:
+- Precision: 72.73%
+- Recall: 88.89%
+- F1-Score: 80.00%
 
 ## Glossary [optional]