ELF22: A Context-based Counter Trolling Dataset to Combat Internet Trolls

ELF22 is a troll-response pairwise dataset along with contextual text from posts in the Reddit community.

Dataset description

The dataset consists of 5,535 pairs, together with 1,151 'your response', divided into the three JSON files: train, validation, and test.

Each post contains a subreddit name, a title, a body text, and two target sentences: a troll comment and the following response. Each post includes two labels: Type of Trolls (2 labels) and Counter Response Strategies (7 labels).

We note that the dataset may contain offensive, sexual, or hateful language. The dataset can be remixed, transformed, and built upon its contents, except for tracking and re-identifying authors. The use of the dataset is considered as the user's consent to the ethical guidelines described in our paper.

Type	Train	Validation	Test	Total
Overt	2,331	166	340	2,837
Covert	3,020	407	422	3,849
Engage	2,331	631	1,339	2,817
Ignore	216	10	35	261
Expose	1,131	75	135	1,359
Challenge	818	48	71	937
Critique	340	31	52	423
Mock	592	51	96	739
Reciprocate	120	18	12	150
All	5,351	573	762	6,686

The fields of post in json files are as follows:

Field	Description
Subreddit	a subreddit name, usually after 'r/'
Title	a title text
Post	a body text
Troll	the root comment with thumbs-down
TrollL	a type of Trolls {1: 'overt', 2: 'covert'}
Response	the following comment with thumbs-up
ResponseL	a counter response strategy {1: 'engage', 2: 'ignore', 3: 'expose', 4:'challenge', 5:'critique', 6: 'mock', 7:'reciprocate'}
Flag	an additional phrase corresponding to its ResponseL for task C. counter response generaetion task
Input	an input text (Title + Post + Troll + Response + Flag) for task C.
Output	an output text (Response) for task C.

Reference

@inproceedings{lee2022elf22,
      title={ELF22: A Context-based Counter Trolling Dataset to Combat Internet Trolls},
      author={Huije Lee and Young Ju NA and Hoyun Song and Jisu Shin and Jong C. Park},
      pages={3530--3541},
      year={2022},
      publisher = {European Language Resources Association},
      booktitle = {Proceedings of the 13th Language Resources and Evaluation, {LREC} 2022, Marseille, France, June 20-25, 2022},
      url={http://www.lrec-conf.org/proceedings/lrec2022/pdf/2022.lrec-1.378.pdf}
}

License

These resources are licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.