|
--- |
|
license: mit |
|
task_categories: |
|
- text-generation |
|
language: |
|
- en |
|
size_categories: |
|
- 100K<n<1M |
|
--- |
|
# IEA_Energy_Dataset |
|
|
|
## Dataset Details |
|
|
|
### Dataset Description |
|
|
|
The dataset is energy-related, covering topics of **Oil**, **Coal**, **Wind**, **Hydrogen**, **Bioenergy**, **Electric vehicles**, **Heating**, **Building envelopes**, **Methane abatement** and **Chemicals**. |
|
|
|
|
|
## Dataset Creation |
|
|
|
### Source Data |
|
|
|
The dataset sources are reports from the webiste of [International Energy Agency(IEA)]('https://www.iea.org/'). |
|
|
|
### Data Collection and Processing |
|
|
|
We scraped free open reports from [IEA's website]("https://www.iea.org/analysis"). The reports are all pdf files and then we used the Llama 3 model to extract useful texts.\ |
|
After getting the raw text, we have the two following steps: |
|
1. Remove too short sentences(e.g length < 100) |
|
2. Use [Monocleaner]('https://github.com/bitextor/monocleaner') to detect disfluent sentences. Each sentence will have a score(0~1), and then we set a threshold to filter. |
|
|
|
|
|
|
