Textual Analysis of Tweets Associated with Domestic Violence

Stephanie  Chua; Janice Allison  Sabang; Keng Sheng  Chew; Puteri Nor Ellyza  Nohuddin

doi:10.18502/ijph.v52i11.14039

Stephanie Chua Faculty of Computer Science and Information Technology, Universiti Malaysia Sarawak, Sarawak, Malaysia
Janice Allison Sabang Faculty of Computer Science and Information Technology, Universiti Malaysia Sarawak, Sarawak, Malaysia
Keng Sheng Chew Faculty of Medicine and Health Sciences, Universiti Malaysia Sarawak, Sarawak, Malaysia
Puteri Nor Ellyza Nohuddin Institute of IR 4.0, Universiti Kebangsaan Malaysia, Selangor, Malaysia

DOI: https://doi.org/10.18502/ijph.v52i11.14039

Keywords: Domestic violence; Twitter; Text analysis

Abstract

Background: Domestic violence is a global public health concern as stated by World Health Organization. We aimed to conduct a textual analysis of tweets associated with domestic violence through keyword identification, word trends and word collocations. The data was obtained from Twitter, focusing on publicly available tweets written in English. The objectives are to find out if the identified keywords, word trends and word collocations can help differentiate between domestic violence-related tweets and non-domestic violence-related tweets, as well as, to analyze the textual characteristics of domestic violence-related tweets and non-domestic violence-related tweets.

Methods: Overall, 11,041 tweets were collected using a few keywords over a period of 15 days from 22 March 2021 to 5 April 2021. A text analysis approach was used to discover the most frequent keywords used, the word trends of those keywords and the word collocations of the keywords in differentiating between domestic violence-related or non-domestic violence-related tweets.

Results: Domestic violence-related tweets and non-domestic violence-related tweets had differentiating characteristics, despite sharing several main keywords. In particular, keywords like “domestic”, “violence” and “suicide” featured prominently in domestic-violence related tweets but not in non-domestic violence-related tweets. Significant differences could also be seen in the frequency of keywords and the word trends in the collection of the tweets.

Conclusion: These findings are significant in helping to automate the flagging of domestic-violence related tweets and alert the authorities so that they can take proactive steps such as assisting the victims in getting medical, police and legal help as needed.