Dataset Summary
SID Classification is a Persian (Farsi) dataset designed for the Classification task, specifically targeting document classification of academic articles. It is a component of the FaMTEB (Farsi Massive Text Embedding Benchmark). The dataset was constructed by collecting academic texts from SID (Scientific Information Database – sid.ir), one of Iran’s major platforms for scientific publications. Each document—formed by concatenating an article’s title and abstract—is labeled with one of 8 predefined scientific disciplines from the SID website.
- Language(s): Persian (Farsi)
- Task(s): Classification (Document Classification, Topic Classification)
- Source: Collected from SID (sid.ir)
- Part of FaMTEB: Yes
Supported Tasks and Leaderboards
This dataset is used to evaluate the performance of text embedding models on the Classification task—measuring their ability to assign academic documents to appropriate scientific categories. Results are benchmarked on the Persian MTEB Leaderboard on Hugging Face Spaces (filtered by language: Persian).
Construction
The dataset was built by:
- Crawling the SID academic repository (sid.ir)
- Extracting the title and abstract of each academic article
- Concatenating the title and abstract to form a single input text
- Using the site's predefined 8-category taxonomy to assign a label to each document
This dataset complements the SID Clustering dataset, sharing the same underlying texts but serving a supervised classification task.
Data Splits
As reported in the FaMTEB paper (Table 5):
- Train: 8,712 samples
- Development (Dev): 0 samples
- Test: 3,735 samples