Stress Detection from Social Media Articles: New Dataset Benchmark and Analytical Study

This repository contains the datasets for classification of stress from text-based social media articles from Reddit and Twitter, which were created within the paper titled "Stress Detection from Social Media Articles: New Dataset Benchmark and Analytical Study".

Presentated orally at IEEE WCCI 2022, IJCNN track (DOI = 10.1109/IJCNN55064.2022.9892889)

Overview of the datasets

We construct four high quality datasets using the text articles from Reddit and Twitter. Against each of the articles is a class label with a value of '0' or '1', where '0' specifies a Stress Negative article and '1' specifies a Stress Positive article. Annotation was done using an automated DNN-based strategy highlighted in the aforementioned study.

The description about each of the datasets is given as under:

• Reddit Title: Consists of titles from the articles collected from both stress and non-stress related subreddits from Reddit.
• Reddit Combi: Consists of title and body text combined together to form a single text sequence, collected from both stress and non-stress related subreddits from Reddit.
• Twitter Full: Consists of stress and non-stress related tweets, collected from Twitter.
• Twitter Non-Advert: Consists of the denoised version of the Twitter Full dataset.

The details about the datasets may be directly referred to from the study.

Other Files

Code files for data preprocessing and finetuning DistilBERT on stress detection are also provided. The files are:

• distilbert_train.py: Finetuning script for DistilBERT.
• distilbert_eval.py: Evaluation script for DistilBERT.
• reddit_preprocessing.py: Preprocessing script for Reddit dataset.
• twitter_preprocessing.py: Preprocessing script for Twitter dataset.

The automated annotation code may be provided on request to the authors.

Citation

@inproceedings{rastogi2022stress,
  title={Stress Detection from Social Media Articles: New Dataset Benchmark and Analytical Study},
  author={Rastogi, Aryan and Liu, Qian and Cambria, Erik},
  booktitle={2022 International Joint Conference on Neural Networks (IJCNN)},
  pages={1--8},
  year={2022},
  organization={IEEE}
}