Update README.md
Browse files
README.md
CHANGED
|
@@ -1,7 +1,7 @@
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
configs:
|
| 4 |
-
- config_name:
|
| 5 |
data_files:
|
| 6 |
- path:
|
| 7 |
- main.jsonl.zst
|
|
@@ -16,4 +16,60 @@ configs:
|
|
| 16 |
split: test
|
| 17 |
language:
|
| 18 |
- en
|
| 19 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: apache-2.0
|
| 3 |
configs:
|
| 4 |
+
- config_name: default
|
| 5 |
data_files:
|
| 6 |
- path:
|
| 7 |
- main.jsonl.zst
|
|
|
|
| 16 |
split: test
|
| 17 |
language:
|
| 18 |
- en
|
| 19 |
+
task_categories:
|
| 20 |
+
- text-classification
|
| 21 |
+
- text-generation
|
| 22 |
+
pretty_name: UltraX-Preview Cleanliness
|
| 23 |
+
tags:
|
| 24 |
+
- preprocessing
|
| 25 |
+
- filter
|
| 26 |
+
- quality
|
| 27 |
+
- cleanliness
|
| 28 |
+
- formatting
|
| 29 |
+
---
|
| 30 |
+
# UltraX-Preview Cleanliness Dataset
|
| 31 |
+
|
| 32 |
+
This dataset evaluates the **cleanliness quality** of raw text by comparing text samples before and after the automated cleaning process.
|
| 33 |
+
|
| 34 |
+
The dataset consists of **50,000 non-duplicate rows** sampled from each source configuration of the [openbmb/UltraX-Preview](https://huggingface.co/datasets/openbmb/UltraX-Preview) dataset.
|
| 35 |
+
|
| 36 |
+
Note: Completely unchanged or completely deleted rows were excluded for numerical stability.
|
| 37 |
+
|
| 38 |
+
## Dataset Configurations & Schema
|
| 39 |
+
|
| 40 |
+
### 1. Main Config (`default`)
|
| 41 |
+
Designed for comparing two text strings: raw and clean.
|
| 42 |
+
|
| 43 |
+
#### Fields
|
| 44 |
+
* `uid`: **String**. Unique identifier matching the original `UltraX-Preview` dataset.
|
| 45 |
+
* `text`: **String**. The raw text from the dataset, truncated to a maximum of 2,000 characters.
|
| 46 |
+
* `raw_distance`: **Integer**. Levenshtein edit distance between x = truncated(raw) and y = truncated(clean).
|
| 47 |
+
* `normalized_distance`: **Float**. Levenshtein distance divided by max(|x|,|y|) (or `0` if both strings are empty).
|
| 48 |
+
* `source`: **String**. The original data source.
|
| 49 |
+
|
| 50 |
+
#### Example JSON
|
| 51 |
+
```json
|
| 52 |
+
{
|
| 53 |
+
"uid": "f8c9620fa931425f2b76e03f5b7b65d0",
|
| 54 |
+
"text": "Dunki has been making notable waves since its release, drawing attention for its narrative and characters. The film has garnered attention not only locally but has also resonated with audiences abroad. A recent instance of this affection was witnessed at the Kerala State School Art Festival, where fans showcased their appreciation by dancing to the tune of the‘Lutt Putt Gaya’ song from Dunki.\n\n\n\nYoung talents at Kerala State School Art Festival groove to Shah Rukh Khan starrer Dunki’s ‘Lutt Putt Gaya’ in their traditional attire; watch\n\nThe festivities at the Kerala State School Art Festival reflected the film's influence, with fans expressing their love through lively dance performances.\n\nYoung talents at the\\#Kerala State School Arts Festival groove to\\#LuttPuttGaya! ❤️ The\\#Dunki Fever is spreading like wildfire! ????\n\n|@iamsrk@RedChilliesEnt@KarunaBadwal@pooja_dadlani |\\#ShahRukhKhan [pic.twitter.com/X02ND17eVH](https://pic.twitter.com/X02ND17eVH)\n\n— Shah Rukh Khan Warriors FAN Club (@TeamSRKWarriors) January 9, 2024\n\n\n\n Dunki was featured at Le Grand Rex, the largest cinema in Europe, on Christmas Evening. Even in foreign locales, the film's release was marked by celebration, as an elderly Shah Rukh Khan fan commemorated the occasion by cutting a cake.\n\nDunki boasts an ensemble cast featuring actors such as Boman Irani, Taapsee Pannu, Vicky Kaushal, Vikram Kochhar, and Anil Grover, alongside Shah Rukh Khan. Presented by JIO Studios, Red Chillies Entertainment, and Rajkumar Hirani Films, and produced by Rajkumar Hirani and Gauri Khan, the film, scripted by Abhijat Joshi, Rajkumar Hirani, and Kanika Dhillon, hit theatres on December 21, 2023.\n\nAlso Read: Anil Grover expresses gratitude towards Shah Rukh Khan, Rajkumar Hirani and others for Dunki journey; says, “A big thanks to t",
|
| 55 |
+
"raw_distance": 278,
|
| 56 |
+
"normalized_distance": 0.1374,
|
| 57 |
+
"source": "AICC"
|
| 58 |
+
}
|
| 59 |
+
|
| 60 |
+
```
|
| 61 |
+
|
| 62 |
+
### 2. Trainable Config (`trainable`)
|
| 63 |
+
|
| 64 |
+
In this configuration, the `normalized_distance` is mapped into a continuous **cleanliness score** via the following transformation:
|
| 65 |
+
|
| 66 |
+
$$\text{Score} = \frac{-\ln(\text{normalized\_distance}) - 2.573391}{1.245167}$$
|
| 67 |
+
|
| 68 |
+
A higher cleanliness score indicates that **less of the raw text needed modification** to achieve the cleaned version (i.e., the text was already relatively clean).
|
| 69 |
+
|
| 70 |
+
10% of the dataset was split off as the test set.
|
| 71 |
+
|
| 72 |
+
## Additional Information
|
| 73 |
+
|
| 74 |
+
* **License:** [Apache 2.0](https://www.apache.org/licenses/LICENSE-2.0)
|
| 75 |
+
* **Exclusion Notice:** No original source artifacts from the underlying preview sources are duplicated or included beyond these derived feature metrics and truncated text samples.
|