---
license: cc-by-2.5
---

# Dataset Card for Dataset Name

<!-- Provide a quick summary of the dataset. -->

This is a dataset of arxiv metadata (including Ids, titles, abstracts and categories) for creating triplets anchor-positive-negative for contrastive learning or evaluation.
The dataset consists of 253140 samples, each sample is arxiv paper metadata, and the negatives of different levels of difficulty (21 levels). 
  

## Dataset Details

The dataset consists of 253140 samples, each sample is a tuple of two elements:
1. An ARXIV paper metadata, including its Id, title and abstract and categories.
2. List of 21 Ids of other ARXIV papers. The first 20 Ids are the papers that are 'closest' to the above paper, and sorted from the most to the least similar; the last 21st Id is an Id of a randomly selected paper (not coinciding with Id of the above paper).

The dataset is created from the mirror of arxiv (snapshot version 173) through the following steps:
1. Identified all arxiv categories with a maximum size of 10K papers (i.e. arxiv preprints).
2. Selected all papers that have at least one of the categories identified above. This is the subset of arxiv to deal with: manageably small, yet diverse.
3. For each paper:
3.1. Sorted its categories by size, from smaller to larger.
3.2. Found all other papers that have the closest match by the categories (the closest match is the longest consecutive list of matched categories, starting from the first one).
3.3. Of the found papers, selected 20 closest by Jensen-Shannon distance between the abstracts, and sorted them by the distance. If there were were less than 20 papers, filled to 20 by the last one.
3.4. Added randomly selected paper as 21st.

For more details see: https://arxiv.org/pdf/2407.00923

### Dataset Description

<!-- Provide a longer summary of what this dataset is. -->

Each sample is a tuple, consisting of two elements. 

First element is an arxiv paper, represented by dictionary, with keys: 'id', 'submitter', 'authors', 'title', 'comments', 'journal-ref', 'doi', 'report-no', 'categories', 'license', 'abstract', 'versions', 'update_date', 'authors_parsed'

Second element is a list of ids of negatives. The first 20 Ids are the papers that are 'closest' to the above paper, and sorted from the most to the least similar; the last 21st Id is an Id of a randomly selected paper. 

- **Curated by:** Oleg Vasilyev
- **License:** cc-by-2.5

### Dataset Sources [optional]

<!-- Provide the basic links for the dataset. -->

- **Repository:** [More Information Needed]
- **Paper [optional]:** "Preserving Multilingual Quality While Tuning Query Encoder on English Only", https://arxiv.org/pdf/2407.00923

## Uses

The dataset was created and used in https://arxiv.org/pdf/2407.00923 for evaluating results of tuning query encoder.

Can be used for creating triplets anchor-positive-negative for contrastive learning or evaluation.

## Dataset Structure

<!-- This section provides a description of the dataset fields, and additional information about the dataset structure such as criteria used to create the splits, relationships between data points, etc. -->

[More Information Needed]

## Dataset Creation

Dataset creation is described in https://arxiv.org/pdf/2407.00923

### Source Data

<!-- This section describes the source data (e.g. news text and headlines, social media posts, translated sentences, ...). -->

arxiv (snapshot version 173)

## Citation [optional]

<!-- If there is a paper or blog post introducing the dataset, the APA and Bibtex information for that should go in this section. -->

**BibTeX:**

@misc{vasilyev2024preservingmultilingualqualitytuning,
      title={Preserving Multilingual Quality While Tuning Query Encoder on English Only}, 
      author={Oleg Vasilyev and Randy Sawaya and John Bohannon},
      year={2024},
      eprint={2407.00923},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2407.00923}, 
}

**APA:**

[More Information Needed]

## Glossary [optional]

<!-- If relevant, include terms and calculations in this section that can help readers understand the dataset or dataset card. -->

[More Information Needed]

## Dataset Card Authors [optional]

Oleg Vasilyev

## Dataset Card Contact

vasilyev.oleg@gmail.com

oleg@primer.ai