Logo Lanfrica

edahanoam/HebTeaseSum

Domaine:

natural language processing

Type de record:

dataset
Créateur:
eda
Hôte:
Hebrew summarization dataset based on our paper "Leveraging Digitized Newspapers to Collect Summarization Data in Low-Resource Languages" ## Quick Start🚀 ### Download the data 1. Install requirements: ```pip install -r requirements.txt``` 2. Choose the wanted portion of HebTeaseSum from the metadata/ directory - **singleDoc.jsonl** All summaries corresponding to a single article - **multiDoc.jsonl** Summaries created from multiple related articles - **singleDoc_over50.jsonl** Summaries with at least 50 words, rated as the highest-quality summaries by human annotators - **multiDoc_over50.jsonl** Summaries with at least 50 words, rated as the highest-quality summaries by human annotators 3. Download the raw XML and build the dataset ``` python datacollector/get_dataset.py --in=$JSONl_FILE --out=$FOLDER_FOR_XMLS --timeout=1 --debug ```