Datasets:

dwb2023
/

ragas-usecase-raw-data

dataset_info:
  features:
    - name: page_content
      dtype: string
    - name: metadata
      struct:
        - name: Judge Comments
          dtype: string
        - name: Judge Score
          dtype: string
        - name: Project Name
          dtype: string
        - name: Score
          dtype: string
        - name: row
          dtype: int64
        - name: source
          dtype: string
  splits:
    - name: train
      num_bytes: 16080
      num_examples: 50
  download_size: 9106
  dataset_size: 16080
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

Dataset Card for ragas-usecase-raw-data

Dataset Description

Source data used to generate ragas-golden-testset-personas. Refer to that dataset for additional information.

Dataset Summary

Capturing here due to some initial challenges with the source data leading to downstream issues with synthetic data generation. (Garbage In, Garbage Out...)

Dataset Structure

The dataset consists of records following the standard LangChain Document format:

page_content (string): Raw text of the source record. Preserves original whitespace, punctuation, and line breaks.
metadata (struct): As-is metadata captured at ingestion (not normalized).
- Judge Comments (string): Free-text qualitative notes from a judge/evaluator.
- Judge Score (string): Score exactly as stored in source (e.g., "4/5", "Good", "N/A").
- Project Name (string): Project/entry name from source.
- Score (string): Additional score field (distinct from Judge Score; may differ).
- row (int64): Original source row index (stable pointer back to raw data).
- source (string): Origin identifier (e.g., file name/path/URI).

Notes: Encoding is UTF-8. Scores remain strings by design. Missing/empty fields may occur.

Supported Tasks

This dataset is primarily intended for:

RAGAS test set generation & evaluation (e.g., faithfulness, answer correctness)
Loading vector stores for RAG (retriever benchmarking, recall/precision diagnostics)

Additional Information

Dataset Curators

The source dataset was curated by dwb2023 and the AI Makerspace team (refer to the citation below).

Licensing Information

This dataset is released under the MIT License.

Citation Information

If you use this dataset in your research, please cite the AI Makerspace team. The original notebook and source data is based on their awesome course!