| --- |
| dataset_info: |
| features: |
| - name: sentences |
| dtype: string |
| - name: labels |
| dtype: int64 |
| splits: |
| - name: train |
| num_bytes: 19977137 |
| num_examples: 8712 |
| - name: test |
| num_bytes: 8607911 |
| num_examples: 3735 |
| download_size: 13060346 |
| dataset_size: 28585048 |
| configs: |
| - config_name: default |
| data_files: |
| - split: train |
| path: data/train-* |
| - split: test |
| path: data/test-* |
| --- |
| ## Dataset Summary |
|
|
| **SID Clustering (SIDClustring)** is a Persian (Farsi) dataset created for the **Clustering** task, specifically focusing on grouping academic articles. It is part of the [FaMTEB (Farsi Massive Text Embedding Benchmark)](https://huggingface.co/spaces/mteb/leaderboard). The dataset was constructed from scientific articles available on **SID (Scientific Information Database – sid.ir)**, categorized into 8 distinct domains reflecting academic disciplines. |
|
|
| * **Language(s):** Persian (Farsi) |
| * **Task(s):** Clustering (Document Clustering, Topic Modeling) |
| * **Source:** Crawled from the SID academic publication platform |
| * **Part of FaMTEB:** Yes |
|
|
| ## Supported Tasks and Leaderboards |
|
|
| This dataset is designed to assess the ability of embedding models to perform document clustering—grouping articles into logical scientific categories. Results can be viewed on the [Persian MTEB Leaderboard](https://huggingface.co/spaces/mteb/leaderboard), under the Clustering task. |
|
|
| ## Construction |
|
|
| 1. Articles were collected by crawling the **sid.ir** platform. |
| 2. For each article: |
| - The **title** and **abstract** were extracted. |
| - These were concatenated using two newline characters (`\n\n`) to form the document input. |
| 3. Each document was assigned to one of 8 predefined SID categories. |
| 4. The resulting dataset serves as a benchmark for evaluating unsupervised clustering performance. |
|
|
| ## Data Splits |
|
|
| * **Train:** 8,712 samples |
| * **Development (Dev):** 0 samples |
| * **Test:** 3,735 samples |
|
|