Update README.md

README.md

@@ -60,6 +60,36 @@ Hence, this dataset is designed to support research and development in biomedica
 3. Assessing retrieval capabilities of various RAG (Retrieval-Augmented Generation) frameworks
 4. Supporting research in biomedical ontologies and knowledge graphs
 
+# BiomixQA Dataset
+
+[Previous sections remain unchanged]
+
+## Performance Analysis
+
+We conducted a comprehensive analysis of the performance of three Large Language Models (LLMs) - Llama-2-13b, GPT-3.5-Turbo (0613), and GPT-4 - on the BiomixQA dataset. We compared their performance using both a standard prompt-based approach and our novel Knowledge Graph Retrieval-Augmented Generation (KG-RAG) framework.
+
+### Performance Summary
+
+Table 1: Performance (accuracy) of LLMs on BiomixQA datasets using prompt-based (zero-shot) and KG-RAG approaches (For more details refer [this](https://arxiv.org/abs/2311.17330) paper)
+
+| Model | True/False Dataset |    | MCQ Dataset |    |
+|-------|-------------------:|---:|------------:|---:|
+|       | Prompt-based | KG-RAG | Prompt-based | KG-RAG |
+| Llama-2-13b | 0.89 ± 0.02 | 0.94 ± 0.01 | 0.31 ± 0.03 | 0.53 ± 0.03 |
+| GPT-3.5-Turbo (0613) | 0.87 ± 0.02 | 0.95 ± 0.01 | 0.63 ± 0.03 | 0.79 ± 0.02 |
+| GPT-4 | 0.90 ± 0.02 | 0.95 ± 0.01 | 0.68 ± 0.03 | 0.74 ± 0.03 |
+
+### Key Observations
+
+1. **Consistent Performance Enhancement**: We observed a consistent performance enhancement for all LLM models when using the KG-RAG framework on both True/False and MCQ datasets.
+
+2. **Significant Improvement for Llama-2**: The KG-RAG framework significantly elevated the performance of Llama-2-13b, particularly on the more challenging MCQ dataset. We observed an impressive 71% increase in accuracy, from 0.31 ± 0.03 to 0.53 ± 0.03.
+
+3. **GPT-4 vs GPT-3.5-Turbo on MCQ**: Intriguingly, we observed a small but statistically significant drop in the performance of the GPT-4 model (0.74 ± 0.03) compared to the GPT-3.5-Turbo model (0.79 ± 0.02) on the MCQ dataset when using the KG-RAG framework. This difference was not observed in the prompt-based approach.
+   - Statistical significance: T-test, p-value < 0.0001, t-statistic = -47.7, N = 1000
+
+4. **True/False Dataset Performance**: All models showed high performance on the True/False dataset, with the KG-RAG approach yielding slightly better results across all models.
+
 
 ## Source Data