mehran-sarmadi's picture
Create README.md
48f780a verified
metadata
configs:
  - config_name: default
    data_files:
      - split: test
        path: qrels/test.jsonl
  - config_name: corpus
    data_files:
      - split: corpus
        path: corpus.jsonl
  - config_name: queries
    data_files:
      - split: queries
        path: queries.jsonl

Dataset Summary

CQADupstack-wordpress-Fa is a Persian (Farsi) dataset created for the Retrieval task, focused on identifying duplicate or semantically equivalent questions in the domain of WordPress development. It is a translated version of the WordPress Development StackExchange data from the English CQADupstack dataset and is part of the FaMTEB (Farsi Massive Text Embedding Benchmark).

  • Language(s): Persian (Farsi)
  • Task(s): Retrieval (Duplicate Question Retrieval)
  • Source: Translated from CQADupstack-WordPress (BEIR benchmark) using Google Translate
  • Part of FaMTEB: Yes — under the BEIR-Fa collection

Supported Tasks and Leaderboards

This dataset is designed to test the ability of text embedding models to retrieve semantically similar or duplicate questions from technical user forums. Evaluation results appear on the Persian MTEB Leaderboard on Hugging Face Spaces (filter by language: Persian).

Construction

The dataset was constructed by:

  • Extracting the WordPress subforum data from the English CQADupstack dataset
  • Translating it into Persian using the Google Translate API
  • Preserving original query-positive pairs for Retrieval task evaluation

As noted in the FaMTEB paper, the BEIR-Fa collection (which includes this dataset) was evaluated through:

  • BM25 score comparisons
  • GEMBA-DA framework using LLMs to assess translation accuracy

These validation methods indicated good translation quality overall.

Data Splits

The full CQADupstack-Fa benchmark includes:

  • Train: 0 samples
  • Dev: 0 samples
  • Test: 480,902 samples (across all CQADupstack-Fa datasets)

This WordPress-specific subset includes approximately 49.9k examples. Individual test splits for this sub-dataset are not separately detailed in the FaMTEB paper; refer to the dataset provider or Hugging Face dataset card for exact distribution if needed.