Datasets:

Modalities:
Text
Formats:
parquet
DOI:
Libraries:
Datasets
Dask
License:
File size: 1,481 Bytes
0c00b58
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ff2b2a9
 
 
 
 
 
 
0c00b58
ff2b2a9
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
---
dataset_info:
  features:
  - name: id
    dtype: string
  - name: embedding
    sequence: float64
  splits:
  - name: train
    num_bytes: 751739666430
    num_examples: 243212198
  download_size: 640572858900
  dataset_size: 751739666430
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
license: cc0-1.0
tags:
- openalex
- embeddings
pretty_name: OpenAlex Embeddings
source_dataset:
- openalex
---
# OpenAlex Embeddings
This dataset contains text embeddings of all records in [OpenAlex](https://openalex.org/) with a title or an abstract from the snapshot of 2023-10-20.
The dataset was created for the [FORAS project](https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=494027) to investigate the efficacy of 
different methods of searching in databases of academic publications. All scripts will be available in a [GitHub repository](https://github.com/IDfuse/foras).

## Description of the data
- The dataset has two columns, `id` and `embedding`. The `id` columns contains the OpenAlex identifier of the record. The `embedding` column contains the text embedding,
which is a vector of 384 floats.
- The multilingual embedding model [intfloat/multilingual-e5-small](https://huggingface.co/intfloat/multilingual-e5-small) was used to generate the embeddings. For every
with a title or abstract we generated an embedding of `'query: '` + `title` + `' '` + `abstract`. The model has a maximum token input length of 512 tokens.