| --- |
| configs: |
| - config_name: default |
| data_files: |
| - split: train |
| path: qrels/train.jsonl |
| - 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 |
|
|
| **SciFact-Fa** is a Persian (Farsi) dataset designed for the **Retrieval** task, with a focus on **scientific fact verification**. It is a translated version of the original English **SciFact** 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 (Scientific Fact Verification, Evidence Retrieval) |
| - **Source:** Translated from the English SciFact dataset using Google Translate |
| - **Part of FaMTEB:** Yes — part of the BEIR-Fa collection |
|
|
| ## Supported Tasks and Leaderboards |
|
|
| This dataset evaluates models' ability to **retrieve supporting evidence** from scientific literature abstracts that either **support or refute** given scientific claims. Evaluation results can be compared on the **Persian MTEB Leaderboard** (filter by language: Persian). |
|
|
| ## Construction |
|
|
| - Translated from the **SciFact** dataset using the **Google Translate API** |
| - The original dataset targets **scientific claim verification** and **evidence-based reasoning** |
|
|
| 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:** 6,102 samples |
| - **Dev:** 0 samples |
| - **Test:** 5,522 samples |
|
|
| > Approximate total dataset size: **7.55k examples** (user-provided figure) |
|
|
