offenseval2020_tr / README.md
albertvillanova's picture
Replace YAML keys from int to str (#2)
43d001a
metadata
annotations_creators:
  - found
language_creators:
  - found
language:
  - tr
license:
  - cc-by-2.0
multilinguality:
  - monolingual
size_categories:
  - 10K<n<100K
source_datasets:
  - original
task_categories:
  - text-classification
task_ids: []
pretty_name: OffensEval-TR 2020
tags:
  - offensive-language-classification
dataset_info:
  features:
    - name: id
      dtype: int32
    - name: tweet
      dtype: string
    - name: subtask_a
      dtype:
        class_label:
          names:
            '0': NOT
            '1': 'OFF'
  config_name: offenseval2020-turkish
  splits:
    - name: train
      num_bytes: 4260505
      num_examples: 31756
    - name: test
      num_bytes: 481300
      num_examples: 3528
  download_size: 2048258
  dataset_size: 4741805

Dataset Card for OffensEval-TR 2020

Table of Contents

Dataset Description

Dataset Summary

The file offenseval-tr-training-v1.tsv contains 31,756 annotated tweets.

The file offenseval-annotation.txt contains a short summary of the annotation guidelines.

Twitter user mentions were substituted by @USER and URLs have been substitute by URL.

Each instance contains up to 1 labels corresponding to one of the following sub-task:

  • Sub-task A: Offensive language identification;

Supported Tasks and Leaderboards

The dataset was published on this paper.

Languages

The dataset is based on Turkish.

Dataset Structure

Data Instances

A binary dataset with with (NOT) Not Offensive and (OFF) Offensive tweets.

Data Fields

Instances are included in TSV format as follows:

ID INSTANCE SUBA

The column names in the file are the following:

id tweet subtask_a

The labels used in the annotation are listed below.

Task and Labels

(A) Sub-task A: Offensive language identification

  • (NOT) Not Offensive - This post does not contain offense or profanity.
  • (OFF) Offensive - This post contains offensive language or a targeted (veiled or direct) offense

In our annotation, we label a post as offensive (OFF) if it contains any form of non-acceptable language (profanity) or a targeted offense, which can be veiled or direct.

Data Splits

train test
31756 3528

Dataset Creation

Curation Rationale

[More Information Needed]

Source Data

[More Information Needed]

Initial Data Collection and Normalization

[More Information Needed]

Who are the source language producers?

From tweeter.

Annotations

[More Information Needed]

Annotation process

We describe the labels above in a “flat” manner. However, the annotation process we follow is hierarchical. The following QA pairs give a more flowchart-like procedure to follow

  1. Is the tweet in Turkish and understandable?
    • No: mark tweet X for exclusion, and go to next tweet
    • Yes: continue to step 2
  2. Is the tweet include offensive/inappropriate language?
    • No: mark the tweet non go to step 4
    • Yes: continue to step 3
  3. Is the offense in the tweet targeted?
    • No: mark the tweet prof go to step 4
    • Yes: chose one (or more) of grp, ind, *oth based on the definitions above. Please try to limit the number of labels unless it is clear that the tweet includes offense against multiple categories.
  4. Was the labeling decision difficult (precise answer needs more context, tweets includes irony, or for another reason)?
    • No: go to next tweet
    • Yes: add the label X, go to next tweet

Who are the annotators?

[More Information Needed]

Personal and Sensitive Information

[More Information Needed]

Considerations for Using the Data

Social Impact of Dataset

[More Information Needed]

Discussion of Biases

[More Information Needed]

Other Known Limitations

[More Information Needed]

Additional Information

Dataset Curators

[More Information Needed]

Licensing Information

The annotations are distributed under the terms of Creative Commons Attribution License (CC-BY). Please cite the following paper, if you use this resource.

Citation Information

@inproceedings{coltekin2020lrec,
 author  = {\c{C}\"{o}ltekin, \c{C}a\u{g}r{\i}},
 year  = {2020},
 title  = {A Corpus of Turkish Offensive Language on Social Media},
 booktitle  = {Proceedings of The 12th Language Resources and Evaluation Conference},
 pages  = {6174--6184},
 address  = {Marseille, France},
 url  = {https://www.aclweb.org/anthology/2020.lrec-1.758},
}

Contributions

Thanks to @yavuzKomecoglu for adding this dataset.