arabartsummarization
| Property | Value |
|---|---|
| License | Apache 2.0 |
| Training Data Size | 37.52K rows |
| Validation Data Size | 4.69K rows |
| Language | Arabic |
What is arabartsummarization?
arabartsummarization is a specialized Arabic text summarization model developed by abdalrahmanshahrour. Built on the mBART architecture, it's designed specifically for Modern Standard Arabic (MSA) text processing, capable of generating concise summaries and news headlines from longer Arabic texts. The model has been trained on a substantial dataset of over 42,000 samples, making it robust for various Arabic text summarization tasks.
Implementation Details
The model utilizes a sequence-to-sequence architecture with the following training specifications: learning rate of 5e-05, batch size of 4, and Adam optimizer with carefully tuned parameters. The training process spanned 3 epochs, achieving a final validation loss of 2.3394.
- Implements beam search with 3 beams for generation
- Uses repetition penalty of 3.0 to ensure diverse outputs
- Supports maximum sequence length of 200 tokens
- Incorporates no-repeat-ngram-size of 3 for better quality
Core Capabilities
- Arabic text summarization for news and general content
- News headline generation
- Arabic paraphrasing
- Handles Modern Standard Arabic (MSA)
- Achieves Rouge1 score of 1.142 and RougeL score of 1.124
Frequently Asked Questions
Q: What makes this model unique?
This model specifically targets Arabic language processing, utilizing the powerful mBART architecture with custom preprocessing through ArabertPreprocessor. It's optimized for news summarization and headline generation, making it particularly valuable for Arabic media and content processing.
Q: What are the recommended use cases?
The model is ideal for news organizations requiring automatic headline generation, content aggregators needing Arabic text summarization, and applications requiring Arabic text condensation while maintaining key information integrity. It's particularly suited for processing Modern Standard Arabic content.
