File size: 2,500 Bytes
fda724e
 
c9ece26
 
 
 
 
 
 
 
 
 
fda724e
304bb38
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
---
license: cc-by-sa-4.0
tags:
- sentiment-analysis
- noise-identification
- noisy-text
size_categories:
- 10K<n<100K
language:
- bn
task_categories:
- text-classification
---

This is a multilabel dataset used for Noise Identification purpose in the paper **"A Comparative Analysis of Noise Reduction Methods in Sentiment Analysis on Noisy Bangla Texts"** accepted in *2024 The 9th Workshop on Noisy and User-generated Text (W-NUT) collocated with EACL 2024*.

- Annotated by 4 native Bangla speakers with 90% trustworthiness score.
- Fleiss' Kappa Score: 0.69

## Definition of noise categories
|Type|Definition|
|-----|---------|
|**Local Word**|Any regional words even if there is a spelling error.|
|**Word Misuse**|Wrong use of words or unnecessary repetitions of words.|
|**Context/Word Missing**|Not enough information or missing words.|
|**Wrong Serial**|Wrong order of the words.|
|**Mixed Language**|Words in another language. Foreign words that were adopted into the Bangla language over time are excluded from this type.|
|**Punctuation Error**|Improper placement or missing punctuation. Sentences ending without "।" were excluded from this type.|
|**Spacing Error**|Improper use of white space.|
|**Spelling Error**|Words not following spelling of Bangla Academy Dictionary.|
|**Coined Word**|Emoji, symbolic emoji, link.|
|**Others**|Noises that do not fall into categories mentioned above.|


## Statistics of NC-SentNoB per noise class
|Class|Instances|#Word/Instance|
|---|---|---|
|**Local Word**|2,084 (0.136%)|16.05|
|**Word Misuse**|661 (0.043%)|18.55|
|**Context/Word Missing**|550 (0.036%)|13.19|
|**Wrong Serial**|69 (0.005%)|15.30
|**Mixed Language**|6,267 (0.410%)|17.91
|**Punctuation Error**|5,988 (0.391%)|17.25|
|**Spacing Error**|2,456 (0.161%)|18.78|
|**Spelling Error**|5,817 (0.380%)|17.30|
|**Coined Words**|549 (0.036%|15.45|
|**Others**|1,263 (0.083%)|16.52|

## Heatmap of correlation coefficient
<img src="https://huggingface.co/datasets/ktoufiquee/NC-SentNoB/resolve/main/corr_heatmap.png">

## Citation
If you use the datasets, please cite the following paper:
```
@misc{elahi2024comparative,
      title={A Comparative Analysis of Noise Reduction Methods in Sentiment Analysis on Noisy Bangla Texts}, 
      author={Kazi Toufique Elahi and Tasnuva Binte Rahman and Shakil Shahriar and Samir Sarker and Md. Tanvir Rouf Shawon and G. M. Shahariar},
      year={2024},
      eprint={2401.14360},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}
```