Update README.md
Browse files
README.md
CHANGED
|
@@ -39,9 +39,7 @@ We have released the latest and largest Chinese dataset, ChineseWebText 2.0, whi
|
|
| 39 |
|
| 40 |
We introduce a new toolchain, MDFG-tool (see Figure 1). We begin with the coarse-grained filtering module, which applies rule-based methods to clean the data, focusing on criteria such as text length and sensitive words to ensure data quality. After cleaning, we evaluate the text quality using a BERT-based model. This process generates a quality score, and by selecting an appropriate threshold, we can extract high-quality text data that meets our needs. Next, we use FastText for both single-label and multi-label classification of the cleaned data. Meanwhile, we conduct toxicity assessment. The FastText model is used to filter out toxic content and assign toxicity scores to each text. This scoring system allows researchers to set thresholds for identifying and selecting harmful texts for further training.
|
| 41 |
|
| 42 |
-
<div align=
|
| 43 |
-
<img src=".\structure.png" width="50%" />
|
| 44 |
-
</div>
|
| 45 |
|
| 46 |
## Citation
|
| 47 |
Please cite the paper if you use the data or code in this repo.
|
|
|
|
| 39 |
|
| 40 |
We introduce a new toolchain, MDFG-tool (see Figure 1). We begin with the coarse-grained filtering module, which applies rule-based methods to clean the data, focusing on criteria such as text length and sensitive words to ensure data quality. After cleaning, we evaluate the text quality using a BERT-based model. This process generates a quality score, and by selecting an appropriate threshold, we can extract high-quality text data that meets our needs. Next, we use FastText for both single-label and multi-label classification of the cleaned data. Meanwhile, we conduct toxicity assessment. The FastText model is used to filter out toxic content and assign toxicity scores to each text. This scoring system allows researchers to set thresholds for identifying and selecting harmful texts for further training.
|
| 41 |
|
| 42 |
+
<div align=center><img src="./structure.png" style="zoom:67%;" /></div>
|
|
|
|
|
|
|
| 43 |
|
| 44 |
## Citation
|
| 45 |
Please cite the paper if you use the data or code in this repo.
|