Datasets:

Flaglab
/

academic-knowledge-abstracts-es

dataset_info:
  features:
    - name: titulo
      dtype: string
    - name: resumen
      dtype: string
    - name: facultad
      dtype: string
  splits:
    - name: train
      num_bytes: 13520606
      num_examples: 10021
    - name: validation
      num_bytes: 1707168
      num_examples: 1253
    - name: test
      num_bytes: 1701241
      num_examples: 1253
  download_size: 9560798
  dataset_size: 16929015
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*
      - split: test
        path: data/test-*
license: cc-by-4.0
task_categories:
  - text-classification
language:
  - es
tags:
  - scientific
size_categories:
  - 10K<n<100K

Academic Knowledge Abstracts (Spanish)

This dataset was built by scraping the institutional research repository of Universidad del Rosario (Colombia).
It contains Spanish-language academic abstracts labeled by the faculty of origin, enabling supervised training and evaluation for faculty classification and domain-specific text modeling tasks.

Dataset Summary

The objective of this dataset is to predict the faculty or academic unit of a paper based solely on its abstract.
We collected all available academic papers from the university’s institutional repository, focusing on open-access entries.
From these documents, we extracted:

The abstract (in Spanish)
Authors
Publication year
Faculty / Department

After cleaning and normalization, each entry was labeled according to one of six main faculties:

Faculty	Number of Samples
Business Administration	3,876
Medicine	3,829
Government	2,691
Law	1,660
Economics	1,076
Engineering	115

Total: 13,247 abstracts (after filtering and normalization).

Languages

Spanish (es)

Data Fields

Each entry contains the following fields:

Field	Type	Description
`abstract`	`string`	The paper’s abstract in Spanish.
`faculty`	`string`	One of six normalized faculty labels.
`authors`	`string`	List of authors (if available).
`year`	`int`	Publication year.
`title`	`string`	Title of the academic paper.
`source_url`	`string`	Original URL in the institutional repository.

Data Splits

The dataset is provided with standard splits for supervised learning:

Split	Size
Train	80%
Validation	10%
Test	10%

Use Cases

This dataset can be used for:

Text classification (predicting academic discipline or faculty)
Domain adaptation of Spanish encoders
Representation learning in scientific Spanish corpora
Educational analytics and institutional knowledge modeling

Ethical Considerations

All texts were collected from open-access academic publications under public institutional repositories.
The dataset contains no personal or sensitive information beyond authorship metadata.

Users should cite the source and respect institutional licenses when reusing or redistributing the dataset.

License

CC BY 4.0 — Attribution required for redistribution or derivative use.