vibhorag101
commited on
Commit
•
e9728c3
1
Parent(s):
2b41ecd
Update README.md
Browse files
README.md
CHANGED
@@ -20,8 +20,26 @@ dataset_info:
|
|
20 |
num_bytes: 18994182.36412781
|
21 |
num_examples: 46394
|
22 |
download_size: 53587175
|
23 |
-
dataset_size: 94970093
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
24 |
---
|
25 |
# Dataset Card for "phr_suicide_prediction_dataset_clean"
|
26 |
-
|
27 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
20 |
num_bytes: 18994182.36412781
|
21 |
num_examples: 46394
|
22 |
download_size: 53587175
|
23 |
+
dataset_size: 94970093
|
24 |
+
license: mit
|
25 |
+
task_categories:
|
26 |
+
- text-classification
|
27 |
+
language:
|
28 |
+
- en
|
29 |
+
pretty_name: Suicidal Tendency Prediction Dataset
|
30 |
+
size_categories:
|
31 |
+
- 100K<n<1M
|
32 |
---
|
33 |
# Dataset Card for "phr_suicide_prediction_dataset_clean"
|
34 |
+
- The dataset is sourced from Reddit and is available on [Kaggle](https://www.kaggle.com/datasets/nikhileswarkomati/suicide-watch).
|
35 |
+
- The dataset contains text with binary labels for suicide or non-suicide.
|
36 |
+
- The dataset was cleaned and following steps were applied
|
37 |
+
- Converted to lowercase
|
38 |
+
- Removed numbers and special characters.
|
39 |
+
- Removed URLs, Emojis and accented characters.
|
40 |
+
- Removed any word contractions.
|
41 |
+
- Remove any extra white spaces and any extra spaces after a single space.
|
42 |
+
- Removed any consecutive characters repeated more than 3 times.
|
43 |
+
- Tokenised the text, then lemmatized it and then removed the stopwords (excluding not).
|
44 |
+
- The `class_label` column was renamed to `label` for use with trainer API.
|
45 |
+
- The evaluation set had ~23000 samples, while the training set had ~186k samples, i.e. a 80:10:10 (train:test:val) split.
|