Update README.md

README.md

@@ -1,26 +1,81 @@
----
-license: apache-2.0
-dataset_info:
-  features:
-  - name: xml_sha1
-    dtype: string
-  - name: declaration_xml
-    dtype: string
-  - name: declaration_json
-    dtype: string
-  - name: text_index
-    dtype: string
-  - name: text_index_embedding
-    sequence: float64
-  splits:
-  - name: train
-    num_bytes: 438508943
-    num_examples: 10944
-  download_size: 204548460
-  dataset_size: 438508943
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
----
+---
+license: apache-2.0
+dataset_info:
+  features:
+  - name: xml_sha1
+    dtype: string
+  - name: declaration_xml
+    dtype: string
+  - name: declaration_json
+    dtype: string
+  - name: text_index
+    dtype: string
+  - name: text_index_embedding
+    sequence: float64
+  splits:
+  - name: train
+    num_bytes: 438508943
+    num_examples: 10944
+  download_size: 204548460
+  dataset_size: 438508943
+configs:
+- config_name: default
+  data_files:
+  - split: train
+    path: data/train-*
+---
+
+# HATVP Declaration Dataset
+
+This dataset contains declarations from the HATVP (Haute Autorité pour la Transparence de la Vie Publique), converted into a structured format suitable for use in retrieval-augmented generation (RAG) tasks.
+
+## Dataset Information
+
+- **License**: `apache-2.0`
+- **Configurations**: This dataset has one configuration (`default`).
+- **Download size**: ~205 MB
+- **Dataset size**: ~439 MB
+
+## Features
+
+- `xml_sha1`: SHA1 hash of the `declaration_xml`, useful for version comparison. Computed on 24-04-2024.
+- `declaration_xml`: Original XML declaration as provided by the HATVP.
+- `declaration_json`: JSON format of the XML, converted using the XMLTODICT Python library. This format reduces the character count by 25%.
+- `text_index`: Name, surname, and current job title of the declaring person.
+- `text_index_embedding`: Embedding vector from the OpenAI `text-embedding-3-large` model, optimized for French language text.
+
+## Data Splits
+
+- **Train**: 10,944 examples, occupying about 438.51 MB.
+
+## Configuration: Default
+
+- **Data Files**:
+  - `train-*`: Contains the training split data files.
+
+## Usage
+
+This dataset facilitates RAG by using a language model to answer questions about individuals based on their HATVP declarations. It is particularly useful for generating factual answers about public figures.
+
+### Example Usage
+
+```python
+# Example of how to use this dataset with a RAG model
+question = "Où travaille le conjoint de Damien Abad?"
+# Assume `rag_model` is a pre-trained model loaded and ready to answer
+answer = rag_model.answer(question)
+print(answer)  # Outputs: "Le conjoint de Damien Abad travaille au Centre Hospitalier du Haut-Bugey."
+```
+
+## Example Interactions
+
+1. **Question**: "Où travaille le conjoint de Damien Abad?"
+   **Answer**: "Le conjoint de Damien Abad travaille au Centre Hospitalier du Haut-Bugey."
+
+2. **Question**: "Combien est payé Damien Abad en 2019?"
+   **Answer**: "According to the provided context, Damien Abad was paid 71,105 euros in 2019."
+
+3. **Question**: "Le conjoint de Damien Abad est-il un homme ou une femme?"
+   **Answer**: "Le conjoint de Damien Abad est une femme, car il est mentionné que son conjoint est une infirmière."
+
+(Answers vary between French and English depending on the language model used for answering.)