Based on the article
ITRC, IRAN, Tehran
This project involves the collection of two datasets: one from Fars News and the other from Tasnim.
Description of the Fars News Dataset :
| category | Number |
|---|---|
| culture | 6000 |
| sports | 5999 |
| politics | 5994 |
| economy | 5992 |
| social | 5991 |
Description of the Tasnim Dataset :
| category | Number |
|---|---|
| culture | 5750 |
| sports | 5347 |
| politics | 5483 |
| economy | 6095 |
| social | 5161 |
Aggregated dataset used :
Two Deep Learning Pretrain model have been experimented:
| category | Precision | Accuracy |
|---|---|---|
| ParsBERT: Transformer-based Model for Persian Language Understanding | 0.85 | 0.83 |
| Language-agnostic BERT Sentence Embedding (LaBSE) | 0.85 | 0.84 |
Confusion matrix using the best model ParsBERT:
| category | label |
|---|---|
| culture | 0 |
| economy | 1 |
| politics | 2 |
| social | 3 |
| sports | 4 |
Marjan Godarzi
Elham Ghasemi
Gholshid Ranjbaran