Update README.md
Browse files
README.md
CHANGED
@@ -11,14 +11,37 @@ size_categories:
|
|
11 |
## Dataset Description
|
12 |
|
13 |
- **Homepage:**
|
14 |
-
- **Repository:**
|
15 |
-
-
|
|
|
|
|
16 |
- **Leaderboard:**
|
17 |
- **Point of Contact:**
|
18 |
|
19 |
### Dataset Summary
|
20 |
|
21 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
22 |
|
23 |
### Supported Tasks and Leaderboards
|
24 |
|
|
|
11 |
## Dataset Description
|
12 |
|
13 |
- **Homepage:**
|
14 |
+
- **Repository:**
|
15 |
+
- [Toxic-Comment-Detection-BN](https://github.com/imbodrulalam/Toxic-Comment-Detection-BN)
|
16 |
+
- **Paper:**
|
17 |
+
- [Bangla Toxic Comment Classification and Severity Measure Using Deep Learning](https://www.researchgate.net/publication/368895245_Bangla_Toxic_Comment_Classification_and_Severity_Measure_Using_Deep_Learning)
|
18 |
- **Leaderboard:**
|
19 |
- **Point of Contact:**
|
20 |
|
21 |
### Dataset Summary
|
22 |
|
23 |
+
Since the deep learning approach needs a huge number
|
24 |
+
of data for model training so it was a major challenge for
|
25 |
+
us to collect a large amount of data to train our model.
|
26 |
+
Some sample comments that we have collected are given
|
27 |
+
below:
|
28 |
+
ছাগেলর বাƐা ছাগল
|
29 |
+
েদেখ পুড়াই িহজড়ার মেতা েদখেত
|
30 |
+
পাডার েপা পাডা েতাের ময্ানেহােল ডু বাইয়া মারেত পারতাম যিদ
|
31 |
+
We have collected almost 4141 labeled data from the
|
32 |
+
previous work of Bangla toxic comment by Jubaer et al.
|
33 |
+
[6], which are described in table 1. For more data, we have
|
34 |
+
collected a total of 22, 000 comments have been collected
|
35 |
+
from Tiktok, the majority of which are toxic comments.
|
36 |
+
Our experts labeled these comments based on 6 categories
|
37 |
+
that are not mutually exclusive. All the annotators are
|
38 |
+
given clear guidelines on how to rate these comments.
|
39 |
+
The guidelines can be summarized in Table I.
|
40 |
+
![Alt text](Capture.PNG)
|
41 |
+
The annotated comments are cleaned by removing
|
42 |
+
emoticons, unnecessary punctuation marks, characters,
|
43 |
+
digits, and other symbols as they contribute very little
|
44 |
+
to the context of the comments.
|
45 |
|
46 |
### Supported Tasks and Leaderboards
|
47 |
|