Update README.md

README.md

@@ -3,8 +3,40 @@ license: mit
 language:
 - en
 tags:
-- accelerator
+- science
+- accelerator-physics
+- particle-accelerator
 pretty_name: Accel-IR
 size_categories:
 - 1K<n<10K
----
+task_categories:
+- text-retrieval
+- question-answering
+---
+# Accel-IR Gold Standard Dataset
+
+This dataset contains expert-annotated Question-Answer pairs for the Particle Accelerator Domain, as described in the Master's Thesis *"From Dataset to Optimization"* by Qing Dai.
+
+## Dataset Structure
+
+Each row represents a question-chunk pair with the following columns:
+
+- **Question**: The domain-specific question (e.g., about beam diagnostics).
+- **Answer**:The answer to the question.
+- **Question_type**:reasoning/summary/definition/fact.
+- **chunk_text**: The paragraph retrieved from technical documentation.
+- **Expert Annotation**: A 1-5 Likert scale rating by domain experts from PSI:
+  - `1`: Irrelevant
+  - `2`: Partially Irrelevant
+  - `3`: Hard to Decide / Not Sure
+  - `4`: Partially Relevant
+  - `5`: Relevant
+- **Label**: Binary label derived from the annotation (1-Yes/0-No).
+- **Source**: The referenced paper or an IPAC publication.
+- **Specific to paper**:If the question is only answerable by the referenced paper, or it's a general question, i.e., non-specific paper.
+
+**Specific to paper** and **Question_type** serve as metadata, allows researchers to explored retrievers' capability in deeper metadata level. 
+
+## Citation
+If you use this dataset, please cite:
+> Qing Dai, "From Dataset to Optimization: A Benchmarking Framework for Information Retrieval in the Particle Accelerator Domain", Master's Thesis, University of Zurich, 2025.