In this research, we focus on the social phenomenon of suicide. Specifically, we perform social sensing on digital traces obtained from Reddit. We analyze the posts and comments in that are related to suicide. We perform natural language processing to better understand different aspects of human life that relate to suicide.

The poster paper for this project was accepted to the 2018 IEEE International Conference on Big Data.

The dataset consists of 406 submissions dating from July 2010 to November 2018 - of which 161 were deemed relevant to suicide and depression - and 9,010 comments (responding to the 161 submissions), that were crawled from https://www.reddit.com/r/singapore/ using the PRAW API. Queries were informed by previous work in "Tracking suicide risk factors through Twitter in the US" by J. Jashinsky, S.H. Burton, C.L. Hanson, J. West, C. Giraud-Carrier, M.D. Barnes, and T. Argyle (2013).

All data was saved to CSV format and can be found in the data folder included in this repository.

pip install -r requirements.txt

This repository was written in Python 3.7.0 and uses the following libraries heavily:

• Gensim LDA Mallet

  sudo apt-get install default-jdk
  sudo apt-get install ant
  git clone [email protected]:mimno/Mallet.git
  cd Mallet/
  ant
  

• GoogleTranslate (Fix: https://stackoverflow.com/questions/52455774/googletrans-stopped-working-with-error-nonetype-object-has-no-attribute-group)

  pip uninstall googletrans
  git clone https://github.com/BoseCorp/py-googletrans.git
  cd ./py-googletrans
  python setup.py install
  

• Spacy

  pip install spacy
  python3 -m spacy download en
  

The crawl_reddit.py script can:

1. --s: crawl submissions from the Singapore subreddit with terms from a vocabulary file,
2. --l: allow data labelling, and/or
3. --c: crawl for comments on the submissions labeled "relevant"

python crawl_reddit.py [optional flags]

The preprocess.py script prepares the data for topic modeling using Gensim or Scikit-learn

python preprocess.py --gensim --sklearn

Both Gensim and Scikit-learn implementations of LDA topic modeling can be run from their respective notebooks. The output plots from pyLDAvis can be found in the plots folder.

• Links embedded in posts (which domains? what are they about?)
• Sentiment analysis (including emoticons and emojis, accounting for sarcasm)
• POS-tagging
• Social network analysis

urlmarker.py written by @gruber was used to Regex match urls during data cleaning.

The malletmodel2ldamodel() method provided by Roger Mähler is an improvement on Gensim's implementation for converting ldamallet to ldamodel.

Many thanks to Dr Kyong Jin Shim for her superb support and guidance.