| --- |
| language: |
| - fa |
| license: mit |
| task_categories: |
| - text-classification |
| - zero-shot-classification |
| - question-answering |
| dataset_info: |
| features: |
| - name: instruction |
| dtype: string |
| - name: input |
| dtype: string |
| - name: output |
| dtype: string |
| splits: |
| - name: train |
| num_bytes: 3272610 |
| num_examples: 1000 |
| download_size: 1528705 |
| dataset_size: 3272610 |
| configs: |
| - config_name: default |
| data_files: |
| - split: train |
| path: data/train-* |
| --- |
| # Persian Telegram Channel Topic Classification Dataset |
|
|
| ## Dataset Description |
|
|
| This dataset is designed for topic classification of Persian text from Telegram channels. It contains a curated collection of Persian text samples along with their corresponding main topics, created using GPT-4 for topic identification. |
|
|
| ### Dataset Details |
|
|
| - **Name**: Persian Telegram Channel Topic Classification Dataset |
| - **Version**: 1.0 |
| - **Size**: 1000 samples |
| - **Language**: Persian (Farsi) |
| - **Source**: Derived from the [PersianTelegramChannels](https://huggingface.co/datasets/mshojaei77/PersianTelegramChannels) dataset |
| - **Format**: Alpaca-style |
|
|
| ## Usage |
|
|
| This dataset can be used for various natural language processing tasks, including: |
|
|
| 1. Topic Classification: |
| - Train a model to predict the main topic of Persian text inputs. |
| - Use the 'input' as features and 'output' as labels for supervised learning. |
|
|
| 2. Fine-tuning Language Models: |
| - Use the dataset to fine-tune Persian language models for better understanding of Telegram-style text. |
| - The 'instruction', 'input', and 'output' format is suitable for instruction-following fine-tuning. |
|
|
| 3. Named Entity Recognition (NER): |
| - Analyze the 'input' texts to identify and extract named entities related to the topics in 'output'. |
| - This can help in building Persian NER models focused on Telegram content. |
|
|
| 4. Multilingual Models: |
| - Incorporate this dataset when training or evaluating multilingual models that include Persian. |
|
|
| 5. Benchmarking: |
| - Use as a benchmark dataset for evaluating various NLP models' performance on Persian text understanding and classification. |
|
|
| Each of these tasks can leverage the Alpaca-style format of the dataset, using the 'instruction', 'input', and 'output' fields as appropriate for the specific NLP task and model architecture. |
|
|
|
|
| ## Dataset Structure |
|
|
| Each entry in the dataset follows the Alpaca-style format: |
|
|
| ```json |
| { |
| "instruction": "موضوع اصلی این متن فارسی را شناسایی کنید", |
| "input": "<Persian text from Telegram channel>", |
| "output": "<Main topic identified by GPT-4>" |
| } |
| ``` |
|
|
| ## Data Fields |
|
|
| - `instruction`: The task instruction (in Persian) |
| - `input`: The original Persian text from a Telegram channel |
| - `output`: The main topic of the text (in Persian) |
|
|
| ## Ethical Considerations |
|
|
| - The dataset contains text from public Telegram channels. Care should be taken to respect privacy and copyright concerns. |
| - The topics were identified by an AI model, which may introduce biases or errors. Users should be aware of potential inaccuracies. |
|
|
| ## Citation |
|
|
| If you use this dataset in your research, please cite: |
|
|
| ``` |
| @dataset{persian_telegram_topic_classification, |
| author = {Mohammad Shojaei}, |
| title = {Persian Telegram Channel Topic Classification Dataset}, |
| year = {2024}, |
| url = {https://huggingface.co/datasets/mshojaei77/alpaca_persian_telegram} |
| } |
| ``` |
