Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,29 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
# WikiBert2WikiBert
|
| 2 |
Bert language models can be employed for Summarization tasks. WikiBert2WikiBert is an encoder-decoder transformer model that is initialized using the Persian WikiBert Model weights. The WikiBert Model is a Bert language model which is fine-tuned on Persian Wikipedia. After using the WikiBert weights for initialization, the model is trained for five epochs on PN-summary and Persian BBC datasets.
|
| 3 |
|
|
@@ -33,27 +59,10 @@ input = 'your input comes here'
|
|
| 33 |
summary = generate_summary(input)
|
| 34 |
```
|
| 35 |
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
task_categories:
|
| 44 |
-
- summarization
|
| 45 |
-
- text generation
|
| 46 |
-
task_ids:
|
| 47 |
-
- news-articles-summarization
|
| 48 |
-
license:
|
| 49 |
-
- apache-2.0
|
| 50 |
-
multilinguality:
|
| 51 |
-
- monolingual
|
| 52 |
-
datasets:
|
| 53 |
-
- pn-summary
|
| 54 |
-
- XL-Sum
|
| 55 |
-
metrics:
|
| 56 |
-
- rouge-1
|
| 57 |
-
- rouge-2
|
| 58 |
-
- rouge-l
|
| 59 |
-
---
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language:
|
| 3 |
+
- fa
|
| 4 |
+
tags:
|
| 5 |
+
- Wikipedia
|
| 6 |
+
- Summarizer
|
| 7 |
+
- bert2bert
|
| 8 |
+
task_categories:
|
| 9 |
+
- summarization
|
| 10 |
+
- text generation
|
| 11 |
+
task_ids:
|
| 12 |
+
- news-articles-summarization
|
| 13 |
+
license:
|
| 14 |
+
- apache-2.0
|
| 15 |
+
multilinguality:
|
| 16 |
+
- monolingual
|
| 17 |
+
datasets:
|
| 18 |
+
- pn-summary
|
| 19 |
+
- XL-Sum
|
| 20 |
+
metrics:
|
| 21 |
+
- rouge-1
|
| 22 |
+
- rouge-2
|
| 23 |
+
- rouge-l
|
| 24 |
+
---
|
| 25 |
+
|
| 26 |
+
|
| 27 |
# WikiBert2WikiBert
|
| 28 |
Bert language models can be employed for Summarization tasks. WikiBert2WikiBert is an encoder-decoder transformer model that is initialized using the Persian WikiBert Model weights. The WikiBert Model is a Bert language model which is fine-tuned on Persian Wikipedia. After using the WikiBert weights for initialization, the model is trained for five epochs on PN-summary and Persian BBC datasets.
|
| 29 |
|
|
|
|
| 59 |
summary = generate_summary(input)
|
| 60 |
```
|
| 61 |
|
| 62 |
+
## Evaluation
|
| 63 |
+
I separated 5 percent of the pn-summary for evaluation of the model. The rouge scores of the model are as follows:
|
| 64 |
+
|
| 65 |
+
| Rouge-1 | Rouge-2 | Rouge-l |
|
| 66 |
+
| ------------- | ------------- | ------------- |
|
| 67 |
+
| 38.97% | 18.42% | 34.50% |
|
| 68 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|