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
SCIDOCS-Fa is a Persian (Farsi) dataset designed for the Retrieval task, with a focus on citation prediction in scientific documents. It is a translated version of the original English SCIDOCS dataset used in the BEIR benchmark and is part of the FaMTEB (Farsi Massive Text Embedding Benchmark) under the BEIR-Fa collection.
- Language(s): Persian (Farsi)
- Task(s): Retrieval (Citation Prediction, Scientific Document Retrieval)
- Source: Translated from the English SCIDOCS dataset using Google Translate
- Part of FaMTEB: Yes — part of the BEIR-Fa collection
Supported Tasks and Leaderboards
This dataset evaluates models' ability to predict citation links by retrieving scientific articles cited by a given citing article, based only on its title. Evaluation results can be compared on the Persian MTEB Leaderboard (filter by language: Persian).
Construction
- Translated from the SCIDOCS dataset using the Google Translate API
- The original dataset focuses on scientific document understanding and citation-based retrieval
Translation quality was validated using:
- BM25 score comparisons
- LLM-based assessment (GEMBA-DA framework)
Data Splits
As reported in the FaMTEB paper (Table 5):
- Train: 0 samples
- Dev: 0 samples
- Test: 30,585 samples
Approximate total dataset size: 56.6k examples (user-provided figure)