README.md

metadata

language:
  - en
multilinguality:
  - monolingual
size_categories:
  - 100M<n<1B
task_categories:
  - feature-extraction
  - sentence-similarity
pretty_name: S2ORC
tags:
  - sentence-transformers
dataset_info:
  - config_name: abstract-citation-pair
    features:
      - name: abstract
        dtype: string
      - name: citation
        dtype: string
    splits:
      - name: train
        num_bytes: 92216608962
        num_examples: 39567485
    download_size: 54303161925
    dataset_size: 92216608962
  - config_name: title-abstract-pair
    features:
      - name: title
        dtype: string
      - name: abstract
        dtype: string
    splits:
      - name: train
        num_bytes: 30708996393
        num_examples: 41769185
    download_size: 19187786420
    dataset_size: 30708996393
  - config_name: title-citation-pair
    features:
      - name: title
        dtype: string
      - name: citation
        dtype: string
    splits:
      - name: train
        num_bytes: 9567159942
        num_examples: 51030086
    download_size: 7054217221
    dataset_size: 9567159942
configs:
  - config_name: abstract-citation-pair
    data_files:
      - split: train
        path: abstract-citation-pair/train-*
  - config_name: title-abstract-pair
    data_files:
      - split: train
        path: title-abstract-pair/train-*
    default: true
  - config_name: title-citation-pair
    data_files:
      - split: train
        path: title-citation-pair/train-*

Dataset Card for S2ORC

This dataset contains titles, abstracts, and citations from scientific papers from the Semantic Scholar Open Research Corpus (S2ORC). This dataset can and has been used to train embedding models, and works out of the box to train or finetune Sentence Transformer models.

In our experiments, title-abstract pairs result in the highest performance, followed by titles-citations and then abstract-citations pairs.

Dataset Subsets

title-abstract-pair subset

• Columns: "title", "abstract"
• Column types: str, str
• Examples:

  {
    "title": "Syntheses, Structures and Properties of Two Transition Metal-Flexible Ligand Coordination Polymers",
    "abstract": "Two coordination polymers based on 3,5-bis(4-carboxyphenylmethyloxy) benzoic acid (H3L), [M(HL)]·2H2O M = Mn(1), Co(2), have been synthesized under hydrothermal conditions. Their structures have been determined by single-crystal X-ray diffraction and further characterized by elemental analysis, IR spectra and TGA. The two complexes possess 3D framework with diamond channels resulting from the trans-configuration of the flexible ligand and three coordination modes, 3(η2, η1), 2(η1, η1), η1, of carboxyl groups in the ligand. The framework can be represented with Schlafli symbol of (48·66)(47·66). The wall of the channel consists of left- or right-handed helical polymeric chains. UV–visible–NIR and photoluminescence spectra, magnetic properties of 1 and 2 have also been discussed.",
  }
  

• Collection strategy: Reading the S2ORC titles-abstract dataset from embedding-training-data.
• Deduplified: No

title-citation-pair subset

• Columns: "title", "citation"
• Column types: str, str
• Examples:

  {
    "title": "An apparent neuroleptic malignant syndrome without extrapyramidal symptoms upon initiation of clozapine therapy: report of a case and results of a clozapine rechallenge.",
    "citation": "Antipsychotic Rechallenge After Neuroleptic Malignant Syndrome with Catatonic Features"
  }
  

• Collection strategy: Reading the S2ORC titles-citation dataset from embedding-training-data and considering each title together with the first citation as a sample.
• Deduplified: No

abstract-citation-pair subset

• Columns: "abstract", "citation"
• Column types: str, str
• Examples:

  {
    "abstract": "The androgen receptor (AR) is a ligand-regulated transcription factor that stimulates cell growth and differentiation in androgen-responsive tissues. The AR N terminus contains two activation functions (AF-1a and AF-1b) that are necessary for maximal transcriptional enhancement by the receptor; however, the mechanisms and components regulating AR transcriptional activation are not fully understood. We sought to identify novel factors that interact with the AR N terminus from an androgen-stimulated human prostate cancer cell library using a yeast two-hybrid approach designed to identify proteins that interact with transcriptional activation domains. A 157-amino acid protein termed ART-27 was cloned and shown to interact predominantly with the AR153–336, containing AF-1a and a part of AF-1b, localize to the nucleus and increase the transcriptional activity of AR when overexpressed in cultured mammalian cells. ART-27 also enhanced the transcriptional activation by AR153–336 fused to the LexA DNA-binding domain but not other AR N-terminal subdomains, suggesting that ART-27 exerts its effect via an interaction with a defined region of the AR N terminus. ART-27 interacts with AR in nuclear extracts from LNCaP cells in a ligand-independent manner. Interestingly, velocity gradient sedimentation of HeLa nuclear extracts suggests that native ART-27 is part of a multiprotein complex. ART-27 is expressed in a variety of human tissues, including sites of androgen action such as prostate and skeletal muscle, and is conserved throughout evolution. Thus, ART-27 is a novel cofactor that interacts with the AR N terminus and plays a role in facilitating receptor-induced transcriptional activation.",
    "citation": "Androgen-insensitivity syndromes in 46,XY fetuses result in various degrees of impairment in genital virilization.1 These syndromes are caused by mutations in the androgen receptor gene that result in decreased binding of androgen to the receptor.2–9 As a consequence, the transcriptional activity of the androgen–androgen-receptor complex is reduced, and therefore, genital virilization is reduced. The androgen receptor, like other steroid hormone receptors, has two major transactivation domains10 — activation function 1 (AF-1) in the N-terminal region11–13 and activation function 2 (AF-2) in the C-terminal ligand-binding domain14 — that interact with the target genes directly as well as indirectly by . . .",
  }
  

• Collection strategy: Reading the S2ORC abstract-citation dataset from embedding-training-data and considering each citation together with the first abstract as a sample.
• Deduplified: No