ACLSum: A New Dataset for Aspect-based Summarization of Scientific Publications
Project description
ACLSum: A New Dataset for Aspect-based Summarization of Scientific Publications
This repository contains data for our paper "ACLSum: A New Dataset for Aspect-based Summarization of Scientific Publications" and a small utility class to work with it.
HuggingFace datasets
You can also use Huggin Face datasets to load ACLSum (dataset link). This would be convenient if you want to train transformer models using our dataset.
Just do,
from datasets import load_dataset
dataset = load_dataset("sobamchan/aclsum")
Our utility class
If you want to see what's in our data more carefully, the following example code on how to use our utility class may be helpful.
You can install the library with the dataset via pip, just run,
pip install aclsum
then you can load the dataset from your python code as,
from aclsum import ACLSum
# Load per split ("train", "val", "test")
train = ACLSum("train")
# One data sample (= paper)
document = train[0]
# Three summaries on each aspect (dict[aspect, summary])
document.summaries
# Get all the sentences from the paper (we only work with abstract, introduction, and conclusion sections) (list[str])
document.get_all_sentences()
# You can specify sections to extract sentences from
document.get_all_sentences(["abstract", "conclusion"])
# Get highlight labels (list[0 or 1])
document.get_all_highlights()
# Get highlighted sentences (list[str])
document.get_all_highlighted_sentences()
Get original PDF parses
While not all the texts are included in the final dataset (only Abstract, Introduction, and Conclusion are included), you can also get the raw output data from Grobid as following,
# This will load a json file in our repo.
raw_data_from_grobid_in_dict = document.get_fulltext_parse()
# For instance you can get author information
raw_data_from_grobid_in_dict["authors"]
# Or the fulltext including other sections
# This will return a list of dicts, {"text": str, "cite_spans": list, "eq_spans": list, "section": str, "sec_num": str}
raw_data_from_grobid_in_dict["pdf_parse"]["body_text"]
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
File details
Details for the file aclsum-0.1.2.tar.gz
.
File metadata
- Download URL: aclsum-0.1.2.tar.gz
- Upload date:
- Size: 5.0 MB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: pdm/2.15.4 CPython/3.10.8 Darwin/22.4.0
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 39fd602a036ab304651a5dd0f1ffdbf58579d2e9c669ae5ebef552f59bd85965 |
|
MD5 | 9149881809e1765cbc3050026d9bbd85 |
|
BLAKE2b-256 | cfdf7a59eb2f3ceca862a3bbb21722cfbadc4e6f76bc1220d614de884541265d |
File details
Details for the file aclsum-0.1.2-py3-none-any.whl
.
File metadata
- Download URL: aclsum-0.1.2-py3-none-any.whl
- Upload date:
- Size: 5.2 MB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: pdm/2.15.4 CPython/3.10.8 Darwin/22.4.0
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | c75ffc48d3eabc5f37c68ea087943780a85950087957759e89746f21111f2213 |
|
MD5 | 8006363afa381fcf36d32e3c6c990177 |
|
BLAKE2b-256 | 55fa636d77b88139b7856c9e980a99a35298c420ad12ac90912fa9f83f9d147a |