README.md

A dataset for benchmarking keyphrase extraction and generation techniques from long document English scientific papers. For more details about the dataset please refer the original paper - .

Data source -

Dataset Summary

Dataset Structure

Data Fields

• id: unique identifier of the document.
• sections: list of all the sections present in the document.
• sec_text: list of white space separated list of words present in each section.
• sec_bio_tags: list of BIO tags of white space separated list of words present in each section.
• extractive_keyphrases: List of all the present keyphrases.
• abstractive_keyphrase: List of all the absent keyphrases.

Data Splits

Split	#datapoints
Train-Small	20,000
Train-Medium	50,000
Train-Large	90,019
Test	3413
Validation	3339

Usage

Small Dataset

from datasets import load_dataset

# get small dataset
dataset = load_dataset("midas/ldkp3k", "small")

def order_sections(sample):
  """
  corrects the order in which different sections appear in the document.
  resulting order is: title, abstract, other sections in the body
  """
  sections = []
  sec_text = []
  sec_bio_tags = []

  if "title" in sample["sections"]:
    title_idx = sample["sections"].index("title")
    sections.append(sample["sections"].pop(title_idx))
    sec_text.append(sample["sec_text"].pop(title_idx))
    sec_bio_tags.append(sample["sec_bio_tags"].pop(title_idx))

  if "abstract" in sample["sections"]:
    abstract_idx = sample["sections"].index("abstract")
    sections.append(sample["sections"].pop(abstract_idx))
    sec_text.append(sample["sec_text"].pop(abstract_idx))
    sec_bio_tags.append(sample["sec_bio_tags"].pop(abstract_idx))

  sections += sample["sections"]
  sec_text += sample["sec_text"]
  sec_bio_tags += sample["sec_bio_tags"]

  return sections, sec_text, sec_bio_tags

# sample from the train split
print("Sample from train data split")
train_sample = dataset["train"][0]

sections, sec_text, sec_bio_tags = order_sections(train_sample)
print("Fields in the sample: ", [key for key in train_sample.keys()])
print("Section names: ", sections)
print("Tokenized Document: ", sec_text)
print("Document BIO Tags: ", sec_bio_tags)
print("Extractive/present Keyphrases: ", train_sample["extractive_keyphrases"])
print("Abstractive/absent Keyphrases: ", train_sample["abstractive_keyphrases"])
print("\n-----------\n")

# sample from the validation split
print("Sample from validation data split")
validation_sample = dataset["validation"][0]

sections, sec_text, sec_bio_tags = order_sections(validation_sample)
print("Fields in the sample: ", [key for key in validation_sample.keys()])
print("Section names: ", sections)
print("Tokenized Document: ", sec_text)
print("Document BIO Tags: ", sec_bio_tags)
print("Extractive/present Keyphrases: ", validation_sample["extractive_keyphrases"])
print("Abstractive/absent Keyphrases: ", validation_sample["abstractive_keyphrases"])
print("\n-----------\n")

# sample from the test split
print("Sample from test data split")
test_sample = dataset["test"][0]

sections, sec_text, sec_bio_tags = order_sections(test_sample)
print("Fields in the sample: ", [key for key in test_sample.keys()])
print("Section names: ", sections)
print("Tokenized Document: ", sec_text)
print("Document BIO Tags: ", sec_bio_tags)
print("Extractive/present Keyphrases: ", test_sample["extractive_keyphrases"])
print("Abstractive/absent Keyphrases: ", test_sample["abstractive_keyphrases"])
print("\n-----------\n")

Output

Medium Dataset

from datasets import load_dataset

# get medium dataset
dataset = load_dataset("midas/ldkp3k", "medium")

Large Dataset

from datasets import load_dataset

# get large dataset
dataset = load_dataset("midas/ldkp3k", "large")

Citation Information

Please cite the works below if you use this dataset in your work.

@article{dl4srmahata2022ldkp,
  title={LDKP - A Dataset for Identifying Keyphrases from Long Scientific Documents},
  author={Mahata, Debanjan and Agarwal, Naveen and Gautam, Dibya and Kumar, Amardeep and Parekh, Swapnil and Singla, Yaman Kumar and Acharya, Anish and Shah, Rajiv Ratn},
  journal={DL4SR-22: Workshop on Deep Learning for Search and Recommendation, co-located with the 31st ACM International Conference on Information and Knowledge Management (CIKM)},
  address={Atlanta, USA},
  month={October},
  year={2022}
}

@article{mahata2022ldkp,
  title={LDKP: A Dataset for Identifying Keyphrases from Long Scientific Documents},
  author={Mahata, Debanjan and Agarwal, Naveen and Gautam, Dibya and Kumar, Amardeep and Parekh, Swapnil and Singla, Yaman Kumar and Acharya, Anish and Shah, Rajiv Ratn},
  journal={arXiv preprint arXiv:2203.15349},
  year={2022}
}

@article{lo2019s2orc,
  title={S2ORC: The semantic scholar open research corpus},
  author={Lo, Kyle and Wang, Lucy Lu and Neumann, Mark and Kinney, Rodney and Weld, Dan S},
  journal={arXiv preprint arXiv:1911.02782},
  year={2019}
}

@inproceedings{ccano2019keyphrase,
  title={Keyphrase generation: A multi-aspect survey},
  author={{\c{C}}ano, Erion and Bojar, Ond{\v{r}}ej},
  booktitle={2019 25th Conference of Open Innovations Association (FRUCT)},
  pages={85--94},
  year={2019},
  organization={IEEE}
}

@article{meng2017deep,
  title={Deep keyphrase generation},
  author={Meng, Rui and Zhao, Sanqiang and Han, Shuguang and He, Daqing and Brusilovsky, Peter and Chi, Yu},
  journal={arXiv preprint arXiv:1704.06879},
  year={2017}
}

Contributions

Thanks to @debanjanbhucs, @dibyaaaaax, @UmaGunturi and @ad6398 for adding this dataset