Update README.md

README.md

@@ -20,14 +20,17 @@ Compared to other ColBERT-based models tested (`colbert-ir/colbertv2.0` and `ans
 ## Model Comparison
 The [AutoRAG Benchmark](https://github.com/Marker-Inc-Korea/AutoRAG-example-korean-embedding-benchmark) serves as both the evaluation dataset and the toolkit for reporting these metrics.
 
-| Model                                   | F1     | Recall | Precision | MAP    | MRR     | NDCG    | Is Best at top_k=3? |
-|------------------------------------------|--------|--------|-----------|--------|---------|---------|--------------------|
-| colbert-ir/colbertv2.0                  | 0.3596 | 0.7193 | 0.2398    | 0.2398 | 0.4459  | 0.5158  | False              |
-| answerai/answerai-colbert-small-v1      | 0.3596 | 0.7193 | 0.2398    | 0.2398 | 0.4240  | 0.4992  | False              |
-| sigridjineth/colbert-small-korean-20241212 | 0.3596 | 0.7193 | 0.2398    | 0.2398 | **0.5278** | **0.5769** | **True**           |
-
-**Key Insight:**  
-- While all three models reach a similar F1 score at `top_k=3`, the `sigridjineth/colbert-small-korean-20241212` model achieves substantially higher MRR and NDCG, indicating better quality rankings and relevance ordering.
+| Model                                     | top_k | F1     | MRR     | NDCG    | Notes                                                                           |
+|-------------------------------------------|-------|--------|---------|---------|---------------------------------------------------------------------------------|
+| colbert-ir/colbertv2.0                    | 1     | 0.2456 | 0.2456  | 0.2456  | Low initial performance.                                                        |
+|                                           | 3     | 0.3596 | 0.4459  | 0.5158  | Shows notable improvement at top_k=3.                                           |
+|                                           | 5     | 0.3596 | 0.4459  | 0.5158  | Similar to top_k=3, no further MRR/NDCG gains.                                  |
+| answerai/answerai-colbert-small-v1        | 1     | 0.2193 | 0.2193  | 0.2193  | Lower performance at top_k=1.                                                   |
+|                                           | 3     | 0.3596 | 0.4240  | 0.4992  | Improved performance at top_k=3, but MRR/NDCG still behind colbertv2.0.          |
+|                                           | 5     | 0.3596 | 0.4240  | 0.4992  | Same as top_k=3, no additional metrics gain.                                    |
+| sigridjineth/colbert-small-korean-20241212| 1     | 0.3772 | 0.3772  | 0.3772  | Highest F1 at top_k=1 among the three models.                                   |
+|                                           | 3     | 0.3596 | **0.5278** | **0.5769** | Slight F1 drop vs. top_k=1, but MRR/NDCG significantly surpass both competitors. |
+|                                           | 5     | 0.3596 | 0.5278  | 0.5769  | Same as top_k=3, maintaining high MRR/NDCG.                                     |
 
 ## Usage
 
@@ -41,7 +44,7 @@ pip install --upgrade colbert-ai
 pip install --upgrade rerankers[transformers]
 ```
 
-### Using `rerankers`
+### Using rerankers
 
 ```python
 from rerankers import Reranker
@@ -52,7 +55,7 @@ query = '센과 치히로의 행방불명을 누가 감독했나요?'
 ranked_docs = ranker.rank(query=query, docs=docs)
 ```
 
-### Using `RAGatouille`
+### Using AGatouille
 
 ```python
 from ragatouille import RAGPretrainedModel
@@ -117,5 +120,4 @@ If you use this model or other JaColBERTv2.5-based models, please cite:
   journal={arXiv preprint arXiv:2407.20750},
   year={2024}
 }
-```
 ```